ARM/decompressor: deal with disabled CP15 barrier instructions
cover ARM/decompressor: deal with disabled CP15 barrier instructions
[ Resending because --cc-cover didn't take ]
While trying to test my v4.4 backport of the firmware/hypervisor based spectre v1/v2 mitigations for 32-bit ARM, I noticed that KVM/qemu failed to boot my kernel while it booted fine under TCG emulation.
As it turns out, KVM/qemu may instantiate the VCPU with support for CP15 barrier instructions disabled, causing them to UNDEF and crash the decompressor.
I already fixed the same issue for UEFI boot, but since v4.4 does not support that, I only noticed now that this is an issue for bare metal as well.
Ard Biesheuvel (2):
ARM/decompressor: avoid CP15 barrier instructions in v7 cache setup
code
Revert "ARM: 8857/1: efi: enable CP15 DMB instructions before cleaning
the cache"
arch/arm/boot/compressed/head.S | 51 ++++++++++---------- 1 file changed, 26 insertions(+), 25 deletions(-)
Signature · 7 lines
2.17.1 _______________________________________________ linux-arm-kernel mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
PATCH RESEND 1/2 ARM/decompressor: avoid CP15 barrier instructions in v7 cache setup code
Commit e17b1af96b2afc38e684aa2f1033387e2ed10029
"ARM: 8857/1: efi: enable CP15 DMB instructions before cleaning the cache"
added some explicit handling of the CP15BEN bit in the SCTLR system register, to ensure that CP15 barrier instructions are enabled, even if we enter the decompressor via the EFI stub.
However, as it turns out, there are other ways in which we may end up using CP15 barrier instructions without them being enabled. I.e., if the decompressor startup code skips the cache_on() call initially, we end up calling cache_clean_flush() with the caches and MMU off, in which case the CP15BEN bit in SCTLR may not be programmed either. And in fact, cache_on() itself issues CP15 barrier instructions before actually enabling them by programming the new SCTLR value (and issuing an ISB)
Since all these routines are specific to v7, let's clean this up by using the ordinary v7 barrier instructions in the v7 specific cache handling routines, so that we never rely on the CP15 ones. This also avoids the issue where a barrier is required between programming SCTLR and using the CP15 barrier instructions, which would result in two different kinds of barriers being used in the same function.
--- arch/arm/boot/compressed/head.S | 35 ++++++++++++++------ 1 file changed, 25 insertions(+), 10 deletions(-)
arch/arm/boot/compressed/head.S +25 -10
@@ -656,6 +656,21 @@ params: ldr r0, =0x10000100 @ params_phys for RPC .align #endif + .macro v7dsb + ARM( .inst 0xf57ff04f @ v7+ dsb ) + THUMB( dsb ) + .endm + + .macro v7dmb + ARM( .inst 0xf57ff05f @ v7+ dmb ) + THUMB( dmb ) + .endm + + .macro v7isb + ARM( .inst 0xf57ff06f @ v7+ isb ) + THUMB( isb ) + .endm + /* * Turn on the cache. We need to setup some page tables so that we * can have both the I and D caches on.
@@ -827,7 +842,7 @@ __armv7_mmu_cache_on: movne r6, #CB_BITS | 0x02 @ !XN blne __setup_mmu mov r0, #0 - mcr p15, 0, r0, c7, c10, 4 @ drain write buffer + v7dsb @ drain write buffer tst r11, #0xf @ VMSA mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs #endif
@@ -849,11 +864,11 @@ __armv7_mmu_cache_on: mcrne p15, 0, r1, c3, c0, 0 @ load domain access control mcrne p15, 0, r6, c2, c0, 2 @ load ttb control #endif - mcr p15, 0, r0, c7, c5, 4 @ ISB + v7isb mcr p15, 0, r0, c1, c0, 0 @ load control register mrc p15, 0, r0, c1, c0, 0 @ and read it back mov r0, #0 - mcr p15, 0, r0, c7, c5, 4 @ ISB + v7isb mov pc, r12 __fa526_cache_on:
@@ -1154,8 +1169,8 @@ __armv7_mmu_cache_off: mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB #endif mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC - mcr p15, 0, r0, c7, c10, 4 @ DSB - mcr p15, 0, r0, c7, c5, 4 @ ISB + v7dsb + v7isb mov pc, r12 /*
@@ -1218,7 +1233,7 @@ __armv7_mmu_cache_flush: mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D b iflush hierarchical: - mcr p15, 0, r10, c7, c10, 5 @ DMB + v7dmb stmfd sp!, {r0-r7, r9-r11} mrc p15, 1, r0, c0, c0, 1 @ read clidr ands r3, r0, #0x7000000 @ extract loc from clidr
@@ -1232,7 +1247,7 @@ loop1: cmp r1, #2 @ see what cache we have at this level blt skip @ skip if no cache, or just i-cache mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr - mcr p15, 0, r10, c7, c5, 4 @ isb to sych the new cssr&csidr + v7isb @ isb to sych the new cssr&csidr mrc p15, 1, r1, c0, c0, 0 @ read the new csidr and r2, r1, #7 @ extract the length of the cache lines add r2, r2, #4 @ add 4 (line length offset)
@@ -1264,10 +1279,10 @@ finished: mov r10, #0 @ switch back to cache level 0 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr iflush: - mcr p15, 0, r10, c7, c10, 4 @ DSB + v7dsb mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB - mcr p15, 0, r10, c7, c10, 4 @ DSB - mcr p15, 0, r10, c7, c5, 4 @ ISB + v7dsb + v7isb mov pc, lr __armv5tej_mmu_cache_flush:
Signature · 7 lines
2.17.1 _______________________________________________ linux-arm-kernel mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
PATCH RESEND 2/2 Revert "ARM: 8857/1: efi: enable CP15 DMB instructions before cleaning the cache"
In reply to Ard BiesheuvelThis reverts commit e17b1af96b2afc38e684aa2f1033387e2ed10029, which is no longer necessary now that the v7 specific routines take care not to issue CP15 barrier instructions before they are enabled in SCTLR.
--- arch/arm/boot/compressed/head.S | 16 +--------------- 1 file changed, 1 insertion(+), 15 deletions(-)
arch/arm/boot/compressed/head.S +1 -15
@@ -1462,21 +1462,7 @@ ENTRY(efi_stub_entry) @ Preserve return value of efi_entry() in r4 mov r4, r0 - - @ our cache maintenance code relies on CP15 barrier instructions - @ but since we arrived here with the MMU and caches configured - @ by UEFI, we must check that the CP15BEN bit is set in SCTLR. - @ Note that this bit is RAO/WI on v6 and earlier, so the ISB in - @ the enable path will be executed on v7+ only. - mrc p15, 0, r1, c1, c0, 0 @ read SCTLR - tst r1, #(1 << 5) @ CP15BEN bit set? - bne 0f - orr r1, r1, #(1 << 5) @ CP15 barrier instructions - mcr p15, 0, r1, c1, c0, 0 @ write SCTLR - ARM( .inst 0xf57ff06f @ v7+ isb ) - THUMB( isb ) - -0: bl cache_clean_flush + bl cache_clean_flush bl cache_off @ Set parameters for booting zImage according to boot protocol
Signature · 7 lines
2.17.1 _______________________________________________ linux-arm-kernel mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Quoted 25 lines · Ard Biesheuvel
[ Resending because --cc-cover didn't take ]
While trying to test my v4.4 backport of the
firmware/hypervisorbased spectre
v1/v2 mitigations for 32-bit ARM, I noticed that KVM/qemu failed to boot my kernel while it booted fine under TCG emulation.As it turns out, KVM/qemu may instantiate the VCPU with support for CP15 barrier instructions disabled, causing them to UNDEF and crash the decompressor.
I already fixed the same issue for UEFI boot, but since v4.4 does not support that, I only noticed now that this is an issue for bare metal as well.
Cc Russell King <[email protected]>Cc Linus Walleij <[email protected]>Cc Marc Zyngier <[email protected]>Ard Biesheuvel (2):
ARM/decompressor: avoid CP15 barrier instructions in v7 cache setup
code
Revert "ARM: 8857/1: efi: enable CP15 DMB instructions before
cleaning
the cache"
Quoted 2 lines
arch/arm/boot/compressed/head.S | 51 ++++++++++---------- 1 file changed, 26 insertions(+), 25 deletions(-)
For the whole series:
M.
Signature · 6 lines
Jazz is not dead. It just smells funny... _______________________________________________ linux-arm-kernel mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-arm-kernel