From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tom Rini Date: Mon, 18 May 2020 14:30:09 -0400 Subject: [PATCH 4/4] efi_loader: call smp_kick_all_cpus() In-Reply-To: <095013977900c8b0c0fceb36e0379a31@walle.cc> References: <20200514123831.30157-1-michael@walle.cc> <20200514123831.30157-5-michael@walle.cc> <94bef691-5899-8ee5-ce94-a6e5fb98dd87@gmx.de> <095013977900c8b0c0fceb36e0379a31@walle.cc> Message-ID: <20200518183009.GG14425@bill-the-cat> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de On Sat, May 16, 2020 at 05:54:43PM +0200, Michael Walle wrote: > [Also adding Tom Rini as ARM maintainer] > > Am 2020-05-14 22:17, schrieb Alexander Graf: > > On 14.05.20 20:46, Heinrich Schuchardt wrote: > > > On 5/14/20 2:38 PM, Michael Walle wrote: > > > > On some architectures, specifically the layerscape, the > > > > secondary cores > > > > wait for an interrupt before entering the spin-tables. This > > > > applies only > > > > to boards which doesn't have PSCI provided by TF-a and u-boot > > > > does the > > > %s/TF-a/TF-A/, %s/u-boot/U-Boot/ > > > > > > > secondary cores handling. > > > > bootm/booti already call that function for ARM architecture; > > > > also add it > > > > to bootelf before switching to EL2. Additionally, provide a weak noop > > > > function so we don't have to have "#ifdef CONFIG_ARM64" guards. > > > > > > > > Signed-off-by: Michael Walle > > > > --- > > > > common/bootm.c | 9 +++++++++ > > > > lib/efi_loader/efi_setup.c | 6 ++++++ > > > > 2 files changed, 15 insertions(+) > > > > > > > > diff --git a/common/bootm.c b/common/bootm.c > > > > index db4362a643..65adf29329 100644 > > > > --- a/common/bootm.c > > > > +++ b/common/bootm.c > > > > @@ -816,6 +816,15 @@ void __weak switch_to_non_secure_mode(void) > > > > { > > > > } > > > > > > > > +/** > > > > + * smp_kick_all_cpus() - kick all CPUs > > > > + * > > > > + * This routine is overridden by architectures requiring this > > > > feature. > > > > + */ > > > > +void __weak smp_kick_all_cpus(void) > > > > +{ > > > > +} > > > > + > > > > #else /* USE_HOSTCC */ > > > > > > > > #if defined(CONFIG_FIT_SIGNATURE) > > > > diff --git a/lib/efi_loader/efi_setup.c b/lib/efi_loader/efi_setup.c > > > > index 26a7423203..7e5364adc5 100644 > > > > --- a/lib/efi_loader/efi_setup.c > > > > +++ b/lib/efi_loader/efi_setup.c > > > > @@ -132,6 +132,12 @@ efi_status_t efi_init_obj_list(void) > > > > /* Allow unaligned memory access */ > > > > allow_unaligned(); > > > > > > > > + /* > > > > + * Some architectures need to kick secondary cores to enter their > > > > + * spin table. > > > > + */ > > > > + smp_kick_all_cpus(); > > > This will not compile with > > > > > > CONFIG_CMD_BOOTI=n > > > CONFIG_CMD_BOOTM=n > > > CONFIG_CMD_BOOTZ=n > > > > > > Much worse is that in incurs needless overhead on PSCI capable > > platforms. Can we move the smp_kick_all_cpus() to the board or soc > > level of the few systems that use spin tables please? :) > > Ok, I'm not sure where I should put the smp_kick_all_cpus(). In the > bootm/booti path this is done in boot_jump_linux() (unconditionally > in do_nonsec_virt_switch()). Therefore, shouldn't the kick be in > switch_to_non_secure_mode(), too? > > Regarding the overhead, isn't it just called once before starting > the OS? Is that really worth adding yet another weak function or > ifdef guards? Isn't it better to behave like the bootm case? So, I think this shows that there's some consolidation needed as VxWorks and ELF support have also had to solve this problem and are copying around the same code that I believe Alex is saying we should in fact not need on SoCs that handle this via PSCI instead of a spintable? -- Tom -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 659 bytes Desc: not available URL: