> -----Original Message----- > From: Marc Zyngier <maz@xxxxxxxxxx> > Sent: Friday, September 9, 2022 7:08 AM > To: Frank Li <frank.li@xxxxxxx> > Cc: tglx@xxxxxxxxxxxxx; robh+dt@xxxxxxxxxx; > krzysztof.kozlowski+dt@xxxxxxxxxx; shawnguo@xxxxxxxxxx; > s.hauer@xxxxxxxxxxxxxx; kw@xxxxxxxxx; bhelgaas@xxxxxxxxxx; linux- > kernel@xxxxxxxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx; linux-arm- > kernel@xxxxxxxxxxxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx; Peng Fan > <peng.fan@xxxxxxx>; Aisheng Dong <aisheng.dong@xxxxxxx>; > jdmason@xxxxxxxx; kernel@xxxxxxxxxxxxxx; festevam@xxxxxxxxx; dl-linux- > imx <linux-imx@xxxxxxx>; kishon@xxxxxx; lorenzo.pieralisi@xxxxxxx; > ntb@xxxxxxxxxxxxxxx; lznuaa@xxxxxxxxx; imx@xxxxxxxxxxxxxxx; > manivannan.sadhasivam@xxxxxxxxxx > Subject: Re: [EXT] Re: [PATCH v9 2/4] irqchip: Add IMX MU MSI controller > driver > > Caution: EXT Email > > On Thu, 08 Sep 2022 16:35:20 +0100, > Frank Li <frank.li@xxxxxxx> wrote: > > > > > > > +struct imx_mu_msi { > > > > > > + spinlock_t lock; > > > > > > + raw_spinlock_t reglock; > > > > > > > > > > Why two locks? Isn't one enough to protect both MSI allocation > (which > > > > > happens once in a blue moon) and register access? > > > > > > > > [Frank Li] Previously your comment, ask me to use raw_spinlock for > > > > read\write register access. I don't think raw_spinlock is good for > > > > MSI allocation. > > > > > > Why wouldn't it be good enough? I'd really like to know.[Frank Li] ' > > > > [Frank Li] According to my understand, raw_spinlock skip some lockdep > > /debug feature to get better performance, which should be used when > > Frequently call, such as irq handle\polling thread. > > I'm afraid you are terribly misguided. They both have the same debug > features because they are both using the same core implementation, and > the only difference is whether this is preemptible for RT purposes or > not. > > > Spinlock have DEBUG feature to check wrong use lock. Allocate MSI > generally > > only is call once when driver probe. > > Again, you should really read the code and the documentation and stop > making things up. [Frank Li] Thanks. You give me the correct direction. Some stackoverflow's Doc was misleaded. I double checked spin_lock implementation. PREEMPT_RT Kernel map spin_lock to rt_mutex. I am curious why exist spin_lock_irqsave and raw_spin_lock_irqsave before PREEMTP_RT merge into kernel tree. > > > > > The basic principle, lock should be used only when necessary. Access reg > and > > Allocate msi is totally independence events. > > Independent events that do not occur simultaneously. So no harm in > sharing the same lock. > > M. > > -- > Without deviation from the norm, progress is not possible.