Re: [PATCH v2 00/11] arm64: qcom: add and enable SHM Bridge support

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Sep 29, 2023 at 08:56:57PM +0200, Bartosz Golaszewski wrote:
> On Fri, Sep 29, 2023 at 5:29 PM Andrew Halaney <ahalaney@xxxxxxxxxx> wrote:
> >
> > On Thu, Sep 28, 2023 at 11:20:29AM +0200, Bartosz Golaszewski wrote:
> > > From: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > >
> > > This is technically the second iteration of the SHM Bridge enablement on
> > > QCom platforms but in practice it's a complete rewrite.
> > >
> > > During the internal discussion with QCom the requirement has been established
> > > as an SHM Bridge pool manager with the assumption that there will be multiple
> > > users, each with their own pool. The problem with this is that we don't have
> > > many potential users of SHM pools upstream at the moment which was rightfully
> > > pointed out in the reviews under v1 (which even had some unused symbols etc.).
> > >
> > > Moreover: after some investigating, it turns out that there simply isn't any
> > > need for multiple pools as the core SCM only allocates a buffer if given call
> > > requires more than 4 arguments and there are only a handful of SCM calls that
> > > need to pass a physical address to a buffer as argument to the trustzone.
> > >
> > > Additionally all but one SCM call allocate buffers passed to the TZ only for
> > > the duration of the call and then free it right aftr it returns. The one
> > > exception is called once and the buffer it uses can remain in memory until
> > > released by the user.
> > >
> > > This all makes using multiple pools wasteful as most of that memory will be
> > > reserved but remain unused 99% of the time. What we need is a single pool of
> > > SCM memory that deals out chunks of suitable format (coherent and
> > > page-aligned) that fulfills the requirements of all calls.
> > >
> > > As not all architectures support SHM bridge, it makes sense to first unify the
> > > memory operations in SCM calls. All users do some kind of DMA mapping to obtain
> > > buffers, most using dma_alloc_coherent() which impacts performance.
> > >
> > > Genalloc pools are very fast so let's use them instead. Create a custom
> > > allocator that also deals with the mapping and use it across all SCM calls.
> > >
> > > Then once this is done, let's extend the allocator to use the SHM bridge
> > > functionality if available on the given platform.
> > >
> > > While at it: let's create a qcom specific directory in drivers/firmware/ and
> > > move relevant code in there.
> > >
> > > I couldn't test all SCM calls but tested with the inline crypto engine on
> > > sm8550 and sa8775p that runs most of new code paths (with and without SHM
> > > bridge). At least qseecom would need some Tested-by.
> >
> > I have been playing around with this on my x13s (sc8280xp). It seems
> > that efivars works ok (and therefore the qseecom stuff I believe), but
> > with these patches applied I'm getting -22 when loading any firmware.
> >
> > I'll try to dig a little more, but thought I'd report that before I
> > context switch for a little bit.
> >
> >     dmesg | grep "firmware\|-22"
> >     [    0.000000] psci: PSCIv1.1 detected in firmware.
> >     [    2.351999] qcom_scm firmware:scm: SHM Bridge enabled
> >     [    2.353002] qcom_scm firmware:scm: qseecom: found qseecom with version 0x1402000
> >     [    6.727604] systemd[1]: systemd-pcrmachine.service - TPM2 PCR Machine ID Measurement was skipped because of an unmet condition check (ConditionPathExists=/sys/firmware/efi/efivars/StubPcrKernelImage-4a67b082-0a4c-41cf-b6c7-440b29bb8c4f).
> >     [    8.198381] qcom_q6v5_pas 1b300000.remoteproc: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qccdsp8280.mbn
> >     [    8.198418] remoteproc remoteproc1: can't start rproc 1b300000.remoteproc: -22
> >     [    8.407641] qcom_q6v5_pas 3000000.remoteproc: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcadsp8280.mbn
> >     [    8.407742] remoteproc remoteproc0: can't start rproc 3000000.remoteproc: -22
> >     [    8.588496] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [    8.588509] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [    9.392185] ath11k_pci 0006:01:00.0: fw_version 0x110b196e fw_build_timestamp 2022-12-22 12:54 fw_build_id WLAN.HSP.1.1-03125-QCAHSPSWPL_V1_V2_SILICONZ_LITE-3.6510.23
> >     [   10.229367] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   10.229383] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.041385] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.041399] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.070144] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.070160] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.321999] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.322015] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.340989] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.341002] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.576180] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.576198] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.593647] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.593661] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.854212] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.854226] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   11.879925] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   11.879937] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   12.157090] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   12.157106] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >     [   12.183074] adreno 3d00000.gpu: error -22 initializing firmware qcom/sc8280xp/LENOVO/21BX/qcdxkmsuc8280.mbn
> >     [   12.183088] msm_dpu ae01000.display-controller: [drm:adreno_load_gpu [msm]] *ERROR* gpu hw init failed: -22
> >
> > Thanks,
> > Andrew
> 
> Huh remoteproc seems to work fine on sm8550. Can you post the full
> kernel log? Do you know which SCM calls fails?
> 
> Bart

See my response on patch 6, and please let me know if you I messed up
the logic there and need to dig a little more. The log didn't have
anything useful outside what is shown here unfortunately, but I can
gather more if you refute that comment on patch 6.

Thanks,
Andrew




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [Linux for Sparc]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux