From mboxrd@z Thu Jan 1 00:00:00 1970 From: ebiederm@xmission.com (Eric W. Biederman) Subject: Re: [PATCH V6 08/33] csky: Process management and Signal Date: Thu, 27 Sep 2018 21:50:43 +0200 Message-ID: <87h8iaoh2k.fsf@xmission.com> References: Mime-Version: 1.0 Content-Type: text/plain Return-path: In-Reply-To: (Guo Ren's message of "Thu, 27 Sep 2018 22:47:45 +0800") Sender: linux-kernel-owner@vger.kernel.org To: Guo Ren Cc: akpm@linux-foundation.org, arnd@arndb.de, daniel.lezcano@linaro.org, davem@davemloft.net, gregkh@linuxfoundation.org, jason@lakedaemon.net, marc.zyngier@arm.com, mark.rutland@arm.com, mchehab+samsung@kernel.org, peterz@infradead.org, robh@kernel.org, robh+dt@kernel.org, tglx@linutronix.de, linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org, devicetree@vger.kernel.org, green.hu@gmail.com List-Id: linux-arch.vger.kernel.org Guo Ren writes: > --- /dev/null > +++ b/arch/csky/abiv2/fpu.c > +void fpu_fpe(struct pt_regs * regs) > +{ > + int sig; > + unsigned int fesr; > + siginfo_t info; > + > + fesr = mfcr("cr<2, 2>"); > + > + if(fesr & FPE_ILLE){ > + info.si_code = ILL_ILLOPC; > + sig = SIGILL; > + } > + else if(fesr & FPE_IDC){ > + info.si_code = ILL_ILLOPN; > + sig = SIGILL; > + } > + else if(fesr & FPE_FEC){ > + sig = SIGFPE; > + if(fesr & FPE_IOC){ > + info.si_code = FPE_FLTINV; > + } > + else if(fesr & FPE_DZC){ > + info.si_code = FPE_FLTDIV; > + } > + else if(fesr & FPE_UFC){ > + info.si_code = FPE_FLTUND; > + } > + else if(fesr & FPE_OFC){ > + info.si_code = FPE_FLTOVF; > + } > + else if(fesr & FPE_IXC){ > + info.si_code = FPE_FLTRES; > + } > + else { > + info.si_code = NSIGFPE; > + } > + } > + else { > + info.si_code = NSIGFPE; > + sig = SIGFPE; > + } > + info.si_signo = SIGFPE; > + info.si_errno = 0; > + info.si_addr = (void *)regs->pc; > + force_sig_info(sig, &info, current); > +} This use of sending a signal is buggy. It results in undefined values being copied to userspace. Userspace should never be sent NSIGXXX as a si_code. You can use FPE_FLTUNK for this default case. In new code please use force_sig_fault instead of force_sig_info in new code. That saves you the trouble of messing with struct siginfo. Thank you very much, Eric Biederman