| Message ID | 20220525032639.1947280-1-mpe@ellerman.id.au (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
| Series | powerpc: Don't select HAVE_IRQ_EXIT_ON_IRQ_STACK | expand |
| Context | Check | Description |
|---|---|---|
| snowpatch_ozlabs/github-powerpc_selftests | success | Successfully ran 10 jobs. |
| snowpatch_ozlabs/github-powerpc_ppctests | success | Successfully ran 10 jobs. |
| snowpatch_ozlabs/github-powerpc_sparse | success | Successfully ran 4 jobs. |
| snowpatch_ozlabs/github-powerpc_kernel_qemu | fail | boot (pmac32_defconfig, mac99, qemu-system-ppc openbios-ppc, ppc-rootfs.cpio.gz, korg-5.5.0, fedo... failed at step Run qemu-mac99 with korg-5.5.0 build kernel. |
| snowpatch_ozlabs/github-powerpc_clang | success | Successfully ran 7 jobs. |
Le 25/05/2022 à 05:26, Michael Ellerman a écrit : > The HAVE_IRQ_EXIT_ON_IRQ_STACK option tells generic code that irq_exit() > is called while still running on the hard irq stack (hardirq_ctx[] in > the powerpc code). > > Selecting the option means the generic code will *not* switch to the > softirq stack before running softirqs, because the code is already > running on the (mostly empty) hard irq stack. > > But since commit 1b1b6a6f4cc0 ("powerpc: handle irq_enter/irq_exit in > interrupt handler wrappers"), irq_exit() is now called on the regular task > stack, not the hard irq stack. > > That's because previously irq_exit() was called in __do_irq() which is > run on the hard irq stack, but now it is called in > interrupt_async_exit_prepare() which is called from do_irq() constructed > by the wrapper macro, which is after the switch back to the task stack. > > So drop HAVE_IRQ_EXIT_ON_IRQ_STACK from the Kconfig. This will mean an > extra stack switch when processing some interrupts, but should > significantly reduce the likelihood of stack overflow. > > It also means the softirq stack will be used for running softirqs from > other interrupts that don't use the hard irq stack, eg. timer interrupts. One thing I have always wondered: what of the point of having both an hardirq stack and a softirq stack ? __do_IRQ() doesn't switch of hardirq stack if we are on softirq stack. do_softirq() voids if already in interrupts. invoke_softirq() calls do_softirq_own_stack() on task_stack. So we have neither situation where we switch from hardirq stack to softirq stack nor from softirq stack to hardirq stack. So wouldn't it be enough to have only one common irq stack ? christophe
On Wed, 25 May 2022 13:26:39 +1000, Michael Ellerman wrote: > The HAVE_IRQ_EXIT_ON_IRQ_STACK option tells generic code that irq_exit() > is called while still running on the hard irq stack (hardirq_ctx[] in > the powerpc code). > > Selecting the option means the generic code will *not* switch to the > softirq stack before running softirqs, because the code is already > running on the (mostly empty) hard irq stack. > > [...] Applied to powerpc/fixes. [1/1] powerpc: Don't select HAVE_IRQ_EXIT_ON_IRQ_STACK https://git.kernel.org/powerpc/c/1346d00e1bdfd4067f92bc14e8a6131a01de4190 cheers
diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig index f5ed355e75e6..7e7d12ae9ecb 100644 --- a/arch/powerpc/Kconfig +++ b/arch/powerpc/Kconfig @@ -221,7 +221,6 @@ config PPC select HAVE_HARDLOCKUP_DETECTOR_PERF if PERF_EVENTS && HAVE_PERF_EVENTS_NMI && !HAVE_HARDLOCKUP_DETECTOR_ARCH select HAVE_HW_BREAKPOINT if PERF_EVENTS && (PPC_BOOK3S || PPC_8xx) select HAVE_IOREMAP_PROT - select HAVE_IRQ_EXIT_ON_IRQ_STACK select HAVE_IRQ_TIME_ACCOUNTING select HAVE_KERNEL_GZIP select HAVE_KERNEL_LZMA if DEFAULT_UIMAGE
The HAVE_IRQ_EXIT_ON_IRQ_STACK option tells generic code that irq_exit() is called while still running on the hard irq stack (hardirq_ctx[] in the powerpc code). Selecting the option means the generic code will *not* switch to the softirq stack before running softirqs, because the code is already running on the (mostly empty) hard irq stack. But since commit 1b1b6a6f4cc0 ("powerpc: handle irq_enter/irq_exit in interrupt handler wrappers"), irq_exit() is now called on the regular task stack, not the hard irq stack. That's because previously irq_exit() was called in __do_irq() which is run on the hard irq stack, but now it is called in interrupt_async_exit_prepare() which is called from do_irq() constructed by the wrapper macro, which is after the switch back to the task stack. So drop HAVE_IRQ_EXIT_ON_IRQ_STACK from the Kconfig. This will mean an extra stack switch when processing some interrupts, but should significantly reduce the likelihood of stack overflow. It also means the softirq stack will be used for running softirqs from other interrupts that don't use the hard irq stack, eg. timer interrupts. Fixes: 1b1b6a6f4cc0 ("powerpc: handle irq_enter/irq_exit in interrupt handler wrappers") Cc: stable@vger.kernel.org # v5.12+ Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> --- arch/powerpc/Kconfig | 1 - 1 file changed, 1 deletion(-)