From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx0a-001b2d01.pphosted.com ([148.163.156.1]:46108 "EHLO mx0a-001b2d01.pphosted.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728472AbgFHOZD (ORCPT ); Mon, 8 Jun 2020 10:25:03 -0400 Subject: Re: [kvm-unit-tests PATCH v8 03/12] s390x: saving regs for interrupts References: <1591603981-16879-1-git-send-email-pmorel@linux.ibm.com> <1591603981-16879-4-git-send-email-pmorel@linux.ibm.com> From: Pierre Morel Message-ID: Date: Mon, 8 Jun 2020 16:24:55 +0200 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8; format=flowed Content-Language: en-US Content-Transfer-Encoding: 7bit Sender: linux-s390-owner@vger.kernel.org List-ID: To: Thomas Huth , kvm@vger.kernel.org Cc: linux-s390@vger.kernel.org, frankja@linux.ibm.com, david@redhat.com, cohuck@redhat.com On 2020-06-08 11:05, Thomas Huth wrote: > On 08/06/2020 10.12, Pierre Morel wrote: >> If we use multiple source of interrupts, for example, using SCLP >> console to print information while using I/O interrupts, we need >> to have a re-entrant register saving interruption handling. >> >> Instead of saving at a static memory address, let's save the base >> registers, the floating point registers and the floating point >> control register on the stack in case of I/O interrupts >> >> Note that we keep the static register saving to recover from the >> RESET tests. >> >> Signed-off-by: Pierre Morel >> Acked-by: Janosch Frank >> --- >> s390x/cstart64.S | 41 +++++++++++++++++++++++++++++++++++++++-- >> 1 file changed, 39 insertions(+), 2 deletions(-) >> >> diff --git a/s390x/cstart64.S b/s390x/cstart64.S >> index b50c42c..a9d8223 100644 >> --- a/s390x/cstart64.S >> +++ b/s390x/cstart64.S >> @@ -119,6 +119,43 @@ memsetxc: >> lmg %r0, %r15, GEN_LC_SW_INT_GRS >> .endm >> >> +/* Save registers on the stack (r15), so we can have stacked interrupts. */ >> + .macro SAVE_REGS_STACK >> + /* Allocate a stack frame for 15 general registers */ >> + slgfi %r15, 15 * 8 >> + /* Store registers r0 to r14 on the stack */ >> + stmg %r0, %r14, 0(%r15) >> + /* Allocate a stack frame for 16 floating point registers */ >> + /* The size of a FP register is the size of an double word */ >> + slgfi %r15, 16 * 8 >> + /* Save fp register on stack: offset to SP is multiple of reg number */ >> + .irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 >> + std \i, \i * 8(%r15) >> + .endr >> + /* Save fpc, but keep stack aligned on 64bits */ >> + slgfi %r15, 8 >> + efpc %r0 >> + stg %r0, 0(%r15) >> + .endm > > I wonder whether it would be sufficient to only save the registers here > that are "volatile" according to the ELF ABI? ... that would save quite > some space on the stack, I think... OTOH, the old code was also saving > all registers, so maybe that's something for a separate patch later... I don't think so for the general registers The "volatile" registers are lost during a C call, so it is the duty of the caller to save them before the call, if he wants, and this is possible for the programmer or the compiler to arrange that. For interruptions, we steal the CPU with all the registers from the program without warning, the program has no possibility to save them. So we must save all registers for him. For the FP registers, we surely can do something if we establish a usage convention on the floating point. A few tests need hardware floating point. If IRQ speed or stack size becomes an issue we can think about optimizing the floating point usage. > > Acked-by: Thomas Huth > Thanks, Pierre -- Pierre Morel IBM Lab Boeblingen