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: <63105bce-6b8e-4b99-bca1-3741f27ea25a@linaro.org>
Date: Wed, 30 Apr 2025 18:19:26 +0200
From: neil.armstrong@...aro.org
To: Konrad Dybcio <konrad.dybcio@....qualcomm.com>,
 Konrad Dybcio <konradybcio@...nel.org>, Rob Clark <robdclark@...il.com>,
 Sean Paul <sean@...rly.run>, Abhinav Kumar <quic_abhinavk@...cinc.com>,
 Dmitry Baryshkov <lumag@...nel.org>, David Airlie <airlied@...il.com>,
 Simona Vetter <simona@...ll.ch>, Bjorn Andersson <andersson@...nel.org>,
 Rob Herring <robh@...nel.org>, Krzysztof Kozlowski <krzk+dt@...nel.org>,
 Conor Dooley <conor+dt@...nel.org>
Cc: Marijn Suijten <marijn.suijten@...ainline.org>,
 linux-arm-msm@...r.kernel.org, dri-devel@...ts.freedesktop.org,
 freedreno@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
 devicetree@...r.kernel.org, Konrad Dybcio <konrad.dybcio@...aro.org>
Subject: Re: [PATCH RFT v6 2/5] drm/msm/adreno: Add speedbin data for SM8550 /
 A740

On 30/04/2025 17:36, Konrad Dybcio wrote:
> On 4/30/25 4:49 PM, neil.armstrong@...aro.org wrote:
>> On 30/04/2025 15:09, Konrad Dybcio wrote:
>>> On 4/30/25 2:49 PM, neil.armstrong@...aro.org wrote:
>>>> On 30/04/2025 14:35, Konrad Dybcio wrote:
>>>>> On 4/30/25 2:26 PM, neil.armstrong@...aro.org wrote:
>>>>>> Hi,
>>>>>>
>>>>>> On 30/04/2025 13:34, Konrad Dybcio wrote:
>>>>>>> From: Konrad Dybcio <konrad.dybcio@...aro.org>
>>>>>>>
>>>>>>> Add speebin data for A740, as found on SM8550 and derivative SoCs.
>>>>>>>
>>>>>>> For non-development SoCs it seems that "everything except FC_AC, FC_AF
>>>>>>> should be speedbin 1", but what the values are for said "everything" are
>>>>>>> not known, so that's an exercise left to the user..
>>>>>>>
>>>>>>> Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@...aro.org>
>>>>>>> Signed-off-by: Konrad Dybcio <konrad.dybcio@...aro.org>
>>>>>>> ---
>>>>>>>      drivers/gpu/drm/msm/adreno/a6xx_catalog.c | 8 ++++++++
>>>>>>>      1 file changed, 8 insertions(+)
>>>>>>>
>>>>>>> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
>>>>>>> index 53e2ff4406d8f0afe474aaafbf0e459ef8f4577d..61daa331567925e529deae5e25d6fb63a8ba8375 100644
>>>>>>> --- a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
>>>>>>> +++ b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
>>>>>>> @@ -11,6 +11,9 @@
>>>>>>>      #include "a6xx.xml.h"
>>>>>>>      #include "a6xx_gmu.xml.h"
>>>>>>>      +#include <linux/soc/qcom/smem.h>
>>>>>>> +#include <linux/soc/qcom/socinfo.h>
>>>>>>> +
>>>>>>>      static const struct adreno_reglist a612_hwcg[] = {
>>>>>>>          {REG_A6XX_RBBM_CLOCK_CNTL_SP0, 0x22222222},
>>>>>>>          {REG_A6XX_RBBM_CLOCK_CNTL2_SP0, 0x02222220},
>>>>>>> @@ -1431,6 +1434,11 @@ static const struct adreno_info a7xx_gpus[] = {
>>>>>>>              },
>>>>>>>              .address_space_size = SZ_16G,
>>>>>>>              .preempt_record_size = 4192 * SZ_1K,
>>>>>>> +        .speedbins = ADRENO_SPEEDBINS(
>>>>>>> +            { ADRENO_SKU_ID(SOCINFO_FC_AC), 0 },
>>>>>>> +            { ADRENO_SKU_ID(SOCINFO_FC_AF), 0 },
>>>>>>> +            /* Other feature codes (on prod SoCs) should match to speedbin 1 */
>>>>>>
>>>>>> I'm trying to understand this sentence. because reading patch 4, when there's no match
>>>>>> devm_pm_opp_set_supported_hw() is simply never called so how can it match speedbin 1 ?
>>>>>
>>>>> What I'm saying is that all other entries that happen to be possibly
>>>>> added down the line are expected to be speedbin 1 (i.e. BIT(1))
>>>>>
>>>>>> Before this change the fallback was speedbin = BIT(0), but this disappeared.
>>>>>
>>>>> No, the default was to allow speedbin mask ~(0U)
>>>>
>>>> Hmm no:
>>>>
>>>>       supp_hw = fuse_to_supp_hw(info, speedbin);
>>>>
>>>>       if (supp_hw == UINT_MAX) {
>>>>           DRM_DEV_ERROR(dev,
>>>>               "missing support for speed-bin: %u. Some OPPs may not be supported by hardware\n",
>>>>               speedbin);
>>>>           supp_hw = BIT(0); /* Default */
>>>>       }
>>>>
>>>>       ret = devm_pm_opp_set_supported_hw(dev, &supp_hw, 1);
>>>>       if (ret)
>>>>           return ret;
>>>
>>> Right, that's my own code even..
>>>
>>> in any case, the kernel can't know about the speed bins that aren't
>>> defined and here we only define bin0, which doesn't break things
>>>
>>> the kernel isn't aware about hw with bin1 with or without this change
>>> so it effectively doesn't matter
>>
>> But it's regression for the other platforms, where before an unknown SKU
>> mapped to supp_hw=BIT(0)
>>
>> Not calling devm_pm_opp_set_supported_hw() is a major regression,
>> if the opp-supported-hw is present, the OPP will be rejected:
> 
> A comment in patch 4 explains that. We can either be forwards or backwards
> compatible (i.e. accept a limited amount of
> speedbin_in_driver x speedbin_in_dt combinations)

I have a hard time understanding the change, please be much more verbose
in the cover letter and commit messages.

The fact that you do such a large change in the speedbin policy in patch 4
makes it hard to understand why it's needed in the first place.

Finally I'm very concerned that "old" SM8550 DT won't work on new kernels,
this is frankly unacceptable, and this should be addressed in the first
place.

The nvmem situation was much simple, where we considered we added the nvmem
property at the same time as opp-supported-hw in OPPs, but it's no more the
case.

So I think the OPP API should probably be extended to address this situation
first, since if we do not have the opp-supported-hw in OPPs, all OPPs are safe.

So this code:
	count = of_property_count_u32_elems(np, "opp-supported-hw");
	if (count <= 0 || count % levels) {
		dev_err(dev, "%s: Invalid opp-supported-hw property (%d)\n",
			__func__, count);
		return false;
	}
should return true in this specific case, like a supported_hw_failsafe mode.

Neil

> 
> Konrad


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ