| Message ID | 20160329100308.GW19428@n2100.arm.linux.org.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Russell, 2016-03-29 19:03 GMT+09:00 Russell King - ARM Linux <linux@arm.linux.org.uk>: > On Tue, Mar 29, 2016 at 10:38:23AM +0900, Masahiro Yamada wrote: >> The Boot ROM of the UniPhier ARMv7 SoCs sets ACTLR (Auxiliary Control >> Register) to different values for different secure states: >> >> [1] Set ACTLR to 0x41 for Non-secure boot >> [2] Set ACTLR to 0x40 for Secure boot >> >> [1] is okay, but [2] is a problem. Because of commit 1b3a02eb4523 >> ("ARMv7: Check whether the SMP/nAMP mode was already enabled"), >> if bit 6 (SMP bit) is already set, the kernel skips the ACTLR setting. >> In that case, bit 0 (FW bit) is never set, so cache ops is not >> broadcasted, causing a cache coherency problem. >> >> To solve the problem, this commit sets the bit 0 of ACTLR if the bit 4 >> has already been set. This change is harmless for [1] because the >> Boot ROM has already set NSACR (Non-secure Access Control Register) >> bit 18 (NS_SMP bit) before switching to Non-secure state in order to >> allow write access to the ACTLR. > > The test in proc-v7.S is too weak, we should probably tighten it to > prevent these kinds of problems, iow: > > arch/arm/mm/proc-v7.S | 10 +++++----- > 1 file changed, 5 insertions(+), 5 deletions(-) > > diff --git a/arch/arm/mm/proc-v7.S b/arch/arm/mm/proc-v7.S > index 0f8963a7e7d9..6fcaac8e200f 100644 > --- a/arch/arm/mm/proc-v7.S > +++ b/arch/arm/mm/proc-v7.S > @@ -281,12 +281,12 @@ __v7_ca17mp_setup: > bl v7_invalidate_l1 > ldmia r12, {r1-r6, lr} > #ifdef CONFIG_SMP > + orr r10, r10, #(1 << 6) @ Enable SMP/nAMP mode > ALT_SMP(mrc p15, 0, r0, c1, c0, 1) > - ALT_UP(mov r0, #(1 << 6)) @ fake it for UP > - tst r0, #(1 << 6) @ SMP/nAMP mode enabled? > - orreq r0, r0, #(1 << 6) @ Enable SMP/nAMP mode > - orreq r0, r0, r10 @ Enable CPU-specific SMP bits > - mcreq p15, 0, r0, c1, c0, 1 > + ALT_UP(mov r0, r10) @ fake it for UP > + orr r10, r10, r0 @ Set required bits > + teq r10, r0 @ Were they already set? > + mcrne p15, 0, r10, c1, c0, 1 @ No, update register > #endif > b __v7_setup_cont > > > I tested it on some of my Cortex-A9 based boards. (all the combinations of SMP/UP SoC and Secure/Non-secure boot) and your patch worked fine! Could you send it as a patch with git-log, please? Please feel free to add my Tested-by: Masahiro Yamada <yamada.masahiro@socionext.com> I retract my crap patch. Thank you!
diff --git a/arch/arm/mm/proc-v7.S b/arch/arm/mm/proc-v7.S index 0f8963a7e7d9..6fcaac8e200f 100644 --- a/arch/arm/mm/proc-v7.S +++ b/arch/arm/mm/proc-v7.S @@ -281,12 +281,12 @@ __v7_ca17mp_setup: bl v7_invalidate_l1 ldmia r12, {r1-r6, lr} #ifdef CONFIG_SMP + orr r10, r10, #(1 << 6) @ Enable SMP/nAMP mode ALT_SMP(mrc p15, 0, r0, c1, c0, 1) - ALT_UP(mov r0, #(1 << 6)) @ fake it for UP - tst r0, #(1 << 6) @ SMP/nAMP mode enabled? - orreq r0, r0, #(1 << 6) @ Enable SMP/nAMP mode - orreq r0, r0, r10 @ Enable CPU-specific SMP bits - mcreq p15, 0, r0, c1, c0, 1 + ALT_UP(mov r0, r10) @ fake it for UP + orr r10, r10, r0 @ Set required bits + teq r10, r0 @ Were they already set? + mcrne p15, 0, r10, c1, c0, 1 @ No, update register #endif b __v7_setup_cont