- 02 May, 2019 40 commits
-
-
Nicholas Piggin authored
"Reconciling" in terms of interrupt handling, is to bring the soft irq mask state in to synch with the hardware, after an interrupt causes MSR[EE] to be cleared (while the soft mask may be enabled, and hard irqs not marked disabled). General kernel code should not be called while unreconciled, because local_irq_disable, etc. manipulations can cause surprising irq traces, and it's fragile because the soft irq code does not really expect to be called in this situation. When exiting from an interrupt, MSR[EE] is cleared to prevent races, but soft irq state is enabled for the returned-to context, so this is now an unreconciled state. restore_math is called in this state, and that can be ftraced, and the ftrace subsystem disables local irqs. Mark restore_math and its callees as notrace. Restore a sanity check in the soft irq code that had to be disabled for this case, by commit 4da1f792 ("powerpc/64: Disable irq restore warning for now"). Signed-off-by:
Nicholas Piggin <npiggin@gmail.com> Signed-off-by:
Michael Ellerman <mpe@ellerman.id.au>
-
Christophe Leroy authored
This patch drops__irq_offset_value which has not been used since commit 9c4cb825 ("powerpc: Remove use of CONFIG_PPC_MERGE") This removes a sparse warning. Fixes: 9c4cb825 ("powerpc: Remove use of CONFIG_PPC_MERGE") Signed-off-by:
Christophe Leroy <christophe.leroy@c-s.fr> Signed-off-by:
-
Christophe Leroy authored
Compared to ifdefs, IS_ENABLED() provide a cleaner code and allows to detect compilation failure regardless of the selected options. Signed-off-by:
-
Christophe Leroy authored
Use cpu_has_feature() instead of opencoding Use IS_ENABLED() instead of #ifdef for CONFIG_TAU_AVERAGE Signed-off-by:
-
Christophe Leroy authored
Use IS_ENABLED() instead of #ifdefs Signed-off-by:
-
Christophe Leroy authored
CPU_FTR_ALTIVEC is only set when CONFIG_ALTIVEC is selected, so the ifdef is unnecessary. Signed-off-by:
-
Christophe Leroy authored
To avoid ifdefs, define a empty static inline mm_iommu_init() function when CONFIG_SPAPR_TCE_IOMMU is not selected. Signed-off-by:
-
Christophe Leroy authored
To avoid #ifdefs, define an static inline fadump_cleanup() function when CONFIG_FADUMP is not selected Signed-off-by:
-
Christophe Leroy authored
No need to add dummy frames when calling trace_hardirqs_on or trace_hardirqs_off. GCC properly handles empty stacks. In addition, powerpc doesn't set CONFIG_FRAME_POINTER, therefore __builtin_return_address(1..) returns NULL at all time. So the dummy frames are definitely unneeded here. In the meantime, avoid reading memory for loading r1 with a value we already know. Signed-off-by:
-
Christophe Leroy authored
As syscalls are now handled via a fast entry path, syscall related actions can be removed from the generic transfer_to_handler path. Signed-off-by:
-
Christophe Leroy authored
This patch implements a fast entry for syscalls. Syscalls don't have to preserve non volatile registers except LR. This patch then implement a fast entry for syscalls, where volatile registers get clobbered. As this entry is dedicated to syscall it always sets MSR_EE and warns in case MSR_EE was previously off It also assumes that the call is always from user, system calls are unexpected from kernel. Signed-off-by:
-
Christophe Leroy authored
This patch implements a fast entry for syscalls. Syscalls don't have to preserve non volatile registers except LR. This patch then implement a fast entry for syscalls, where volatile registers get clobbered. As this entry is dedicated to syscall it always sets MSR_EE and warns in case MSR_EE was previously off It also assumes that the call is always from user, system calls are unexpected from kernel. The overall series improves null_syscall selftest by 12,5% on an 83xx and by 17% on a 8xx. Signed-off-by:
-
Christophe Leroy authored
[text mostly copied from benh's RFC/WIP] ppc32 are still doing something rather gothic and wrong on 32-bit which we stopped doing on 64-bit a while ago. We have that thing where some handlers "copy" the EE value from the original stack frame into the new MSR before transferring to the handler. Thus for a number of exceptions, we enter the handlers with interrupts enabled. This is rather fishy, some of the stuff that handlers might do early on such as irq_enter/exit or user_exit, context tracking, etc... should be run with interrupts off afaik. Generally our handlers know when to re-enable interrupts if needed. The problem we were having is that we assumed these interrupts would return with interrupts enabled. However that isn't the case. Instead, this patch changes things so that we always enter exception handlers with interrupts *off* with the notable exception of syscalls which are special (and get a fast path). Suggested-by:
Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by:
-
Christophe Leroy authored
EXC_XFER_TEMPLATE() is not called with COPY_EE anymore so we can get rid of copyee parameters and related COPY_EE and NOCOPY macros. Suggested-by:
-
Christophe Leroy authored
All exceptions handlers know when to reenable interrupts, so it is safer to enter all of them with MSR_EE unset, except for syscalls. Suggested-by:
-
Christophe Leroy authored
syscalls are expected to be entered with MSR_EE set. Lets make it inconditional by forcing MSR_EE on syscalls. This patch adds EXC_XFER_SYS for that. Suggested-by:
-
Christophe Leroy authored
SPEFloatingPointException() is the only exception handler which 'forgets' to re-enable interrupts. This patch makes sure it does. Suggested-by:
-
Christophe Leroy authored
Refactor exception entry macros by using the ones defined in head_32.h Signed-off-by:
-
Christophe Leroy authored
This patch splits NORMAL_EXCEPTION_PROLOG in the same way as in head_8xx.S and head_32.S and renames it EXCEPTION_PROLOG() as well to match head_32.h Signed-off-by:
-
Christophe Leroy authored
This patch adds STACK_FRAME_REGS_MARKER in the stack at exception entry in order to see interrupts in call traces as below: [ 0.013964] Call Trace: [ 0.014014] [c0745db0] [c007a9d4] tick_periodic.constprop.5+0xd8/0x104 (unreliable) [ 0.014086] [c0745dc0] [c007aa20] tick_handle_periodic+0x20/0x9c [ 0.014181] [c0745de0] [c0009cd0] timer_interrupt+0xa0/0x264 [ 0.014258] [c0745e10] [c000e484] ret_from_except+0x0/0x14 [ 0.014390] --- interrupt: 901 at console_unlock.part.7+0x3f4/0x528 [ 0.014390] LR = console_unlock.part.7+0x3f0/0x528 [ 0.014455] [c0745ee0] [c0050334] console_unlock.part.7+0x114/0x528 (unreliable) [ 0.014542] [c0745f30] [c00524e0] register_console+0x3d8/0x44c [ 0.014625] [c0745f60] [c0675aac] cpm_uart_console_init+0x18/0x2c [ 0.014709] [c0745f70] [c06614f4] console_init+0x114/0x1cc [ 0.014795] [c0745fb0] [c0658b68] start_kernel+0x300/0x3d8 [ 0.014864] [c0745ff0] [c00022cc] start_here+0x44/0x98 Signed-off-by:
-
Christophe Leroy authored
Unlike said in the comment, r1 is not reused by the critical exception handler, as it uses a dedicated critirq_ctx stack. Decrementing r1 early is then unneeded. Should the above be valid, the code is crap buggy anyway as r1 gets some intermediate values that would jeopardise the whole process (for instance after mfspr r1,SPRN_SPRG_THREAD) Using SPRN_SPRG_SCRATCH2 to save r1 is then not needed, r11 can be used instead. This avoids one mtspr and one mfspr and makes the prolog closer to what's done on 6xx and 8xx. Signed-off-by:
-
Christophe Leroy authored
6xx/8xx EXC_XFER_TEMPLATE() macro adds a i##n symbol which is unused and can be removed. 40x and booke EXC_XFER_TEMPLATE() macros takes msr from the caller while the 6xx/8xx version uses only MSR_KERNEL as msr value. This patch modifies the 6xx/8xx version to make it similar to the 40x and booke versions. Signed-off-by:
-
Christophe Leroy authored
As preparation for using head_32.h for head_40x.S, move LOAD_MSR_KERNEL() there and use it to load r10 with MSR_KERNEL value. In the mean time, this patch modifies it so that it takes into account the size of the passed value to determine if 'li' can be used or if 'lis/ori' is needed instead of using the size of MSR_KERNEL. This is done by using gas macro. Signed-off-by:
-
Christophe Leroy authored
EXCEPTION_PROLOG is similar in head_8xx.S and head_32.S This patch creates head_32.h and moves EXCEPTION_PROLOG macro into it. It also converts it from a GCC macro to a GAS macro in order to ease refactorisation with 40x later, since GAS macros allows the use of #ifdef/#else/#endif inside it. And it also has the advantage of not requiring the uggly "; \" at the end of each line. This patch also moves EXCEPTION() and EXC_XFER_XXXX() macros which are also similar while adding START_EXCEPTION() out of EXCEPTION(). Signed-off-by:
-
Christophe Leroy authored
Reduce #ifdef mess by defining a helper to print hash info at startup. In the meantime, remove the display of hash table address to reduce leak of non necessary information. Signed-off-by:
-
Christophe Leroy authored
Due to %p, (ptrval) is printed in lieu of the hash table address. showing the hash table address isn't an operationnal need so just don't print it. Signed-off-by:
-
Christophe Leroy authored
Hash_end has never been used, drop it. Signed-off-by:
-
Christophe Leroy authored
For hash32, the zero shadow page gets mapped with PAGE_READONLY instead of PAGE_KERNEL_RO, because the PP bits don't provide a RO kernel, so PAGE_KERNEL_RO is equivalent to PAGE_KERNEL. By using PAGE_READONLY, the page is RO for both kernel and user, but this is not a security issue as it contains only zeroes. Signed-off-by:
-
Christophe Leroy authored
KASAN requires early activation of hash table, before memblock() functions are available. This patch implements an early hash_table statically defined in __initdata. During early boot, a single page table is used. For hash32, when doing the final init, one page table is allocated for each PGD entry because of the _PAGE_HASHPTE flag which can't be common to several virt pages. This is done after memblock get available but before switching to the final hash table, otherwise there are issues with TLB flushing due to the shared entries. Signed-off-by:
-
Christophe Leroy authored
For KASAN, hash table handling will be activated early for accessing to KASAN shadow areas. In order to avoid any modification of the hash functions while they are still used with the early hash table, the code patching is moved out of MMU_init_hw() and put close to the big-bang switch to the final hash table. Signed-off-by:
-
Christophe Leroy authored
This patch adds KASAN support for PPC32. The following patch will add an early activation of hash table for book3s. Until then, a warning will be raised if trying to use KASAN on an hash 6xx. To support KASAN, this patch initialises that MMU mapings for accessing to the KASAN shadow area defined in a previous patch. An early mapping is set as soon as the kernel code has been relocated at its definitive place. Then the definitive mapping is set once paging is initialised. For modules, the shadow area is allocated at module_alloc(). Signed-off-by:
-
Christophe Leroy authored
All files containing functions run before kasan_early_init() is called must have KASAN instrumentation disabled. For those file, branch profiling also have to be disabled otherwise each if () generates a call to ftrace_likely_update(). Signed-off-by:
-
Christophe Leroy authored
This patch prepares a shadow area for KASAN. The shadow area will be at the top of the kernel virtual memory space above the fixmap area and will occupy one eighth of the total kernel virtual memory space. Signed-off-by:
-
Christophe Leroy authored
When we add KASAN shadow area, KVIRT_TOP can't be anymore fixed at 0xfe000000. This patch uses FIXADDR_START to define KVIRT_TOP. Signed-off-by:
-
Christophe Leroy authored
Since commit 400c47d8 ("powerpc32: memset: only use dcbz once cache is enabled"), memset() can be used before activation of the cache, so no need to use memset_io() for zeroing the BSS. Acked-by:
Dmitry Vyukov <dvyukov@google.com> Signed-off-by:
-
Christophe Leroy authored
In kernel/cputable.c, explicitly use memcpy() instead of *y = *x; This will allow GCC to replace it with __memcpy() when KASAN is selected. Acked-by:
-
Christophe Leroy authored
When KASAN is active, the string functions in lib/ are doing the KASAN checks. This is too early for prom_init. This patch implements dedicated string functions for prom_init, which will be compiled in with KASAN disabled. Size of prom_init before the patch: text data bss dec hex filename 12060 488 6960 19508 4c34 arch/powerpc/kernel/prom_init.o Size of prom_init after the patch: text data bss dec hex filename 12460 488 6960 19908 4dc4 arch/powerpc/kernel/prom_init.o This increases the size of prom_init a bit, but as prom_init is in __init section, it is freed after boot anyway. Signed-off-by:
-
Christophe Leroy authored
This patch makes CONFIG_CMDLINE defined at all time. It avoids having to enclose related code inside #ifdef CONFIG_CMDLINE Signed-off-by:
-
Christophe Leroy authored
CONFIG_KASAN implements wrappers for memcpy() memmove() and memset() Those wrappers are doing the verification then call respectively __memcpy() __memmove() and __memset(). The arches are therefore expected to rename their optimised functions that way. For files on which KASAN is inhibited, #defines are used to allow them to directly call optimised versions of the functions without going through the KASAN wrappers. See commit 393f203f ("x86_64: kasan: add interceptors for memset/memmove/memcpy functions") for details. Other string / mem functions do not (yet) have kasan wrappers, we therefore have to fallback to the generic versions when KASAN is active, otherwise KASAN checks will be skipped. Signed-off-by:
-
Christophe Leroy authored
In preparation of KASAN, move early_init() into a separate file in order to allow deactivation of KASAN for that function. Signed-off-by:
-
