linux-samsung-soc.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Jaehoon Chung <jh80.chung@samsung.com>
To: Ulf Hansson <ulf.hansson@linaro.org>,
	Marek Szyprowski <m.szyprowski@samsung.com>,
	Krzysztof Kozlowski <krzk@kernel.org>
Cc: Christian Hewitt <christianshewitt@gmail.com>,
	"linux-samsung-soc@vger.kernel.org" 
	<linux-samsung-soc@vger.kernel.org>,
	Marian Mihailescu <mihailescu2m@gmail.com>,
	Sylwester Nawrocki <snawrocki@kernel.org>
Subject: Re: BUG: Cannot boot Odroid XU4 from eMMC without this patch
Date: Thu, 7 Oct 2021 07:12:12 +0900	[thread overview]
Message-ID: <1c345d2e-e028-2a3c-6b38-368841542332@samsung.com> (raw)
In-Reply-To: <CAPDyKFr0gbjzTPs1gxAg=1DPhe7tM6R9sGT9rSCX2BtFh2hekA@mail.gmail.com>

On 10/6/21 7:40 PM, Ulf Hansson wrote:
> On Mon, 4 Oct 2021 at 17:51, Marek Szyprowski <m.szyprowski@samsung.com> wrote:
>>
>> Hi Krzysztof,
>>
>> On 04.10.2021 15:01, Krzysztof Kozlowski wrote:
>>> On 04/10/2021 12:07, Marek Szyprowski wrote:
>>>> On 01.10.2021 02:40, Jaehoon Chung wrote:
>>>>> On 9/27/21 8:09 PM, Krzysztof Kozlowski wrote:
>>>>>> On Mon, 13 Sept 2021 at 06:32, Christian Hewitt
>>>>>> <christianshewitt@gmail.com> wrote:
>>>>>>> https://protect2.fireeye.com/v1/url?k=6f7d4070-30e6793d-6f7ccb3f-0cc47aa8f5ba-2c8976d4b015314f&q=1&e=776d64d2-22f3-400a-a241-42af8b5f60d0&u=https%3A%2F%2Fgithub.com%2Fchewitt%2Flinux%2Fcommit%2F8a4ebfb43a394e5dc5e9fafc92a50d5e81a4f258
>>>>>>>
>>>>>>> If I boot any recent kernel without the above patch, the emmc module on the XU4 is not detected, see:
>>>>>>>
>>>>>>> Without:
>>>>>>>
>>>>>>> [    3.227837] mmc0: tuning execution failed: -5
>>>>>>> [    3.231229] mmc0: error -5 whilst initialising MMC card
>>>>>>> [    3.536450] mmc0: tuning execution failed: -5
>>>>>>> [    3.539680] mmc0: error -5 whilst initialising MMC card
>>>>>>> [    3.794056] mmc0: tuning execution failed: -5
>>>>>>> [    3.794212] mmc0: error -5 whilst initialising MMC card
>>>>>>> [    4.111097] mmc0: tuning execution failed: -5
>>>>>>> [    4.115356] mmc0: error -5 whilst initialising MMC card
>>>>>>> [    4.426164] mmc0: tuning execution failed: -5
>>>>>>> [    4.429678] mmc0: error -5 whilst initialising MMC card
>>>>>>> [    4.756226] mmc0: tuning execution failed: -5
>>>>>>> [    4.760641] mmc0: error -5 whilst initialising MMC card
>>>>>>>
>>>>>>> With:
>>>>>>>
>>>>>>> [    3.305461] mmc0: new HS400 MMC card at address 0001
>>>>>>> [    3.307444] mmcblk0: mmc0:0001 8GME4R 7.28 GiB
>>>>>>> [    3.308132] mmcblk0boot0: mmc0:0001 8GME4R 4.00 MiB
>>>>>>> [    3.309172] mmcblk0boot1: mmc0:0001 8GME4R 4.00 MiB
>>>>>>> [    3.310255] mmcblk0rpmb: mmc0:0001 8GME4R 512 KiB, chardev (246:0)
>>>>>>> [    3.315963]  mmcblk0: p1 p2
>>>>>>>
>>>>>>> The patch is sourced from a Linux 5.4 patchset used by several retro gaming distros for XU4 images shared in the HardKernel forums. I would be happy to submit it, but the original patch has no description in the commit message. Not being a coding developer myself I cannot explain whether it is correct or what it’s doing to add one. All I can do is confirm that it works, and is needed. SD card boot is not an issue.
>>>>>>>
>>>>>>> I’ve CC’d the original author (Marian) in case he remembers the patch and can comment. It would be good to get this upstream.
>>>>>> The patch might have sense but would require describing conditions -
>>>>>> what MMC input and output clock settings work and which do not work.
>>>>>> Also someone would need to test other Exynos5422 boards and other
>>>>>> Exynos with HS200 and HS400 support (Exynos5433, Exynos7). I think
>>>>>> this should not affect SD cards.
>>>>> Thanks for adding me.
>>>>> I didn't see XU4 booting fail with linux-5.15-rc1 kernel.
>>>>>
>>>>> [    4.561934] mmc1: new HS400 MMC card at address 0001
>>>>> [    4.572401] mmcblk1: mmc1:0001 SDW16G 14.7 GiB
>>>>> [    4.602555]  mmcblk1: p1 p2 p3 p4 < p5 p6 p7 >
>>>>> [    4.623201] mmcblk1boot0: mmc1:0001 SDW16G 4.00 MiB
>>>>> [    4.640465] mmcblk1boot1: mmc1:0001 SDW16G 4.00 MiB
>>>>>
>>>>> Which kernel version did you use?
>>>> I came across this patch some time ago, but also didn't manage to
>>>> reproduce the issue - in my case eMMC was always detected properly. It
>>>> might be related to particular version or series of the eMMC modules.
>>>>
>>>> I've just checked that patch on XU3, XU4, PeachPi and TM2e boards. All
>>>> are working properly with it.
>>>>
>>>> I've also tried to benchmark the impact of that change and in some case
>>>> it causes some performance degradation.
>>>>
>>>> The main difference is clock configuration. Before this patch (XU4):
>>>>
>>>> # dmesg | grep mmc0
>>>> mmc_host mmc0: Bus speed (slot 0) = 50000000Hz (slot req 400000Hz,
>>>> actual 396825HZ div = 63)
>>>> mmc_host mmc0: Bus speed (slot 0) = 200000000Hz (slot req 200000000Hz,
>>>> actual 200000000HZ div = 0)
>>>> mmc_host mmc0: Bus speed (slot 0) = 50000000Hz (slot req 52000000Hz,
>>>> actual 50000000HZ div = 0)
>>>> mmc_host mmc0: Bus speed (slot 0) = 400000000Hz (slot req 200000000Hz,
>>>> actual 200000000HZ div = 1)
>>>> mmc0: new HS400 MMC card at address 0001
>>>> mmcblk0: mmc0:0001 SDW16G 14.7 GiB
>>>> mmcblk0boot0: mmc0:0001 SDW16G 4.00 MiB
>>>> mmcblk0boot1: mmc0:0001 SDW16G 4.00 MiB
>>>> mmcblk0rpmb: mmc0:0001 SDW16G 4.00 MiB, chardev (245:0)
>>>>
>>>> After applying the patch (mmc device number is random depending on the
>>>> boot):
>>>>
>>>> # dmesg | grep mmc1
>>>> [    3.619177] mmc_host mmc1: Bus speed (slot 0) = 50000000Hz (slot req
>>>> 400000Hz, actual 396825HZ div = 63)
>>>> [    4.057167] mmc_host mmc1: Bus speed (slot 0) = 200000000Hz (slot req
>>>> 200000000Hz, actual 200000000HZ div = 0)
>>>> [    4.070040] mmc_host mmc1: Bus speed (slot 0) = 50000000Hz (slot req
>>>> 52000000Hz, actual 50000000HZ div = 0)
>>>> [    4.089028] mmc_host mmc1: Bus speed (slot 0) = 266666666Hz (slot req
>>>> 200000000Hz, actual 133333333HZ div = 1)
>>>> [    4.102296] mmc1: new HS400 MMC card at address 0001
>>>> [    4.119072] mmcblk1: mmc1:0001 SDW16G 14.7 GiB
>>>> [    4.173507] mmcblk1boot0: mmc1:0001 SDW16G 4.00 MiB
>>>> [    4.196210] mmcblk1boot1: mmc1:0001 SDW16G 4.00 MiB
>>>> [    4.215163] mmcblk1rpmb: mmc1:0001 SDW16G 4.00 MiB, chardev (245:0)
>>>>
>>>> The performance has been measured with:
>>>>
>>>> # dd if=/dev/mmcblk1p6 of=/dev/null bs=128k
>>>> 31944+0 records in
>>>> 31944+0 records out
>>>> 4186963968 bytes (4.2 GB) copied, 36.6981 s, 114 MB/s
>>>>
>>>> Results (XU4 board):
>>>>
>>>> exynos_defconfig: 145 MB/s (before) vs 114 MB/s (after)
>>>> exynos_defconfig + all devfreqs set to performance: 146 MB/s vs 115 MB/s
>>>> exynos_defconfig + cpufreq & all devfreqs set to performance: 154 MB/s
>>>> vs 139 MB/s
>>>> exynos_defconfig + CONFIG_ARM_EXYNOS_BUS_DEVFREQ disabled: 130 MB/s vs
>>>> 108 MB/s
>>>> exynos_defconfig + CONFIG_CPUFREQ_DT disabled: 69 MB/s (no impact)
>>>> exynos_defconfig + CONFIG_ARM_EXYNOS_BUS_DEVFREQ & CONFIG_CPUFREQ_DT: 66
>>>> MB/s (no impact)
>>>>
>>>>
>>>> Maybe some other clock configuration (I mean the rate of the top-level
>>>> clocks or even PLLS) will solve the issue without degrading the
>>>> performance, but it is hard to judge that without reproducing the issue.
>>>>
>>> Thanks for testing and performance measurements. The read speed seems to
>>> be a direct effect of lower mmc bus frequency (DIVRATIO changed from 2
>>> to 3).
>>>
>>> What next? Except what Marek suggested, maybe compare MMC card
>>> capabilities and try to find a difference? For example Marek's eMMC (at
>>> least one with logs) is from Sandisk SDW16G. Christian's and Marian's
>>> (Memeka / mihailescu2m) is apparently Samsung's 8GME4R.
>>
>> I have also eMMC modules with AJTD4R and 016G92 IDs. I can test them,
>> but that won't happen soon due to remote work.
>>
>>> Because the patch reduces performance and not all users are affected, I
>>> have some doubts. Maybe use by default lower clock (so as in this patch)
>>> but if card is in list of known good cards go higher?
>>>
>>> Another idea is to use always slower bus because it is simply safer and
>>> we do not have resources to test 100x of different eMMC modules...
>>>
>>> Any comments from you?
>>
>> I vote for the safer clock rates. Maybe we can add some dt-property for
>> the higher rates for the boards with the known to be working properly
>> eMMC modules.
> 
> We already have a common mmc DT property, "max-frequency". I think
> that should fit well, as it allows us to cap the frequency on those
> boards that need this, right?

It's already using "max-frequency" for those boards.I think that its value is correct.
I will check this issue on this week. Will share after checking what's best solution.

> 
> Having a card quirk could be another option, but then we need to be
> sure that it's actually the card that has the limitation.

I don't want to use a card quirk if there is no solution to issue.

Best Regards,
Jaehoon Chung

> 
> Kind regards
> Uffe
> 


  reply	other threads:[~2021-10-06 22:11 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-13  4:27 Christian Hewitt
2021-09-27 11:09 ` Krzysztof Kozlowski
2021-10-01  0:40   ` Jaehoon Chung
2021-10-01 13:41     ` Christian Hewitt
2021-10-06 22:16       ` Jaehoon Chung
2021-10-07  4:36         ` Christian Hewitt
2021-10-07 11:05           ` Jaehoon Chung
2021-10-20 11:14             ` Jaehoon Chung
2021-10-04 10:07     ` Marek Szyprowski
2021-10-04 13:01       ` Krzysztof Kozlowski
2021-10-04 15:51         ` Marek Szyprowski
2021-10-06 10:40           ` Ulf Hansson
2021-10-06 22:12             ` Jaehoon Chung [this message]
2021-10-06 22:01       ` Jaehoon Chung

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1c345d2e-e028-2a3c-6b38-368841542332@samsung.com \
    --to=jh80.chung@samsung.com \
    --cc=christianshewitt@gmail.com \
    --cc=krzk@kernel.org \
    --cc=linux-samsung-soc@vger.kernel.org \
    --cc=m.szyprowski@samsung.com \
    --cc=mihailescu2m@gmail.com \
    --cc=snawrocki@kernel.org \
    --cc=ulf.hansson@linaro.org \
    --subject='Re: BUG: Cannot boot Odroid XU4 from eMMC without this patch' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).