linux-kernel - Re: [PATCH 6/6] firmware: qcom_scm: Only compile legacy calls on ARM

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: <161501150705.1478170.3739297122787060750@swboyd.mtv.corp.google.com>
Date:   Fri, 05 Mar 2021 22:18:27 -0800
From:   Stephen Boyd <swboyd@...omium.org>
To:     Andy Gross <agross@...nel.org>,
        Bjorn Andersson <bjorn.andersson@...aro.org>,
        Elliot Berman <eberman@...eaurora.org>
Cc:     linux-kernel@...r.kernel.org, linux-arm-msm@...r.kernel.org,
        Brian Masney <masneyb@...tation.org>,
        Stephan Gerhold <stephan@...hold.net>,
        Jeffrey Hugo <jhugo@...eaurora.org>,
        Douglas Anderson <dianders@...omium.org>
Subject: Re: [PATCH 6/6] firmware: qcom_scm: Only compile legacy calls on ARM

Quoting Elliot Berman (2021-03-05 10:18:09)
> On 3/3/2021 10:14 PM, Stephen Boyd wrote:
> > Quoting Elliot Berman (2021-03-03 19:35:08)
> >>
> >> On 2/23/2021 1:45 PM, Stephen Boyd wrote:
> >>> These scm calls are never used outside of legacy ARMv7 based platforms.
> >>> That's because PSCI, mandated on arm64, implements them for modern SoCs
> >>> via the PSCI spec. Let's move them to the legacy file and only compile
> >>> the legacy file into the kernel when CONFIG_ARM=y. Otherwise provide
> >>> stubs and fail the calls. This saves a little bit of space in an
> >>> arm64 allmodconfig >
> >>>    $ ./scripts/bloat-o-meter vmlinux.before vmlinux.after
> >>>    add/remove: 0/8 grow/shrink: 5/7 up/down: 509/-4405 (-3896)
> >>>    Function                                     old     new   delta
> >>>    __qcom_scm_set_dload_mode.constprop          312     452    +140
> >>>    qcom_scm_qsmmu500_wait_safe_toggle           288     416    +128
> >>>    qcom_scm_io_writel                           288     408    +120
> >>>    qcom_scm_io_readl                            376     492    +116
> >>>    __param_str_download_mode                     23      28      +5
> >>>    __warned                                    4327    4326      -1
> >>>    qcom_iommu_init                              272     268      -4
> >>>    e843419@...f_00010432_324                      8       -      -8
> >>>    qcom_scm_call                                228     208     -20
> >>>    CSWTCH                                      5925    5877     -48
> >>>    _sub_I_65535_1                            163100  163040     -60
> >>>    _sub_D_65535_0                            163100  163040     -60
> >>>    qcom_scm_wb                                   64       -     -64
> >>>    qcom_scm_lock                                320     160    -160
> >>>    qcom_scm_call_atomic                         212       -    -212
> >>>    qcom_scm_cpu_power_down                      308       -    -308
> >>>    scm_legacy_call_atomic                       520       -    -520
> >>>    qcom_scm_set_warm_boot_addr                  720       -    -720
> >>>    qcom_scm_set_cold_boot_addr                  728       -    -728
> >>>    scm_legacy_call                             1492       -   -1492
> >>>    Total: Before=66737642, After=66733746, chg -0.01%
> >>>
> >>> Commit 9a434cee773a ("firmware: qcom_scm: Dynamically support SMCCC and
> >>> legacy conventions") didn't mention any motivating factors for keeping
> >>> the legacy code around on arm64 kernels, i.e. presumably that commit
> >>> wasn't trying to support these legacy APIs on arm64 kernels.
> >>
> >> There are arm targets which support SMCCC convention and use some of
> >> these removed functions. Can these functions be kept in qcom-scm.c and
> >> wrapped with #if IS_ENABLED(CONFIG_ARM)?
> >>
> > 
> > It can be wrapped in qcom-scm.c, but why? It's all the same object file
> > so I'm lost why it matters. I suppose it would make it so the struct
> > doesn't have to be moved around and declared in the header? Any other
> > reason? I moved it to the legacy file so that it was very obvious that
> > the API wasn't to be used except for "legacy" platforms that don't use
> > PSCI.
> > 
> 
> There are "legacy" arm platforms that use the SMCCC (scm_smc_call) and 
> use the qcom_scm_set_{warm,cold}_boot_addr and qcom_scm_cpu_power_down 
> functions.

Ah ok. Weird, but I get it. Amazing that SMCCC was adopted there but
PSCI wasn't!

> 
>  > +    desc.args[0] = flags;
>  > +    desc.args[1] = virt_to_phys(entry);
>  > +
>  > +    return scm_legacy_call_atomic(NULL, &desc, NULL);
>  > +}
>  > +EXPORT_SYMBOL(qcom_scm_set_cold_boot_addr);
> 
> This should still be qcom_scm_call.

You mean s/scm_legacy_call_atomic/qcom_scm_call/ right?

I don't really want to resend the rest of the patches if this last one
is the only one that needs an update. This was a semi-RFC anyway so
maybe it's fine if the first 5 patches get merged and then I can resend
this one? Otherwise I will resend this again next week or so with less
diff for this patch.