[RFC PATCH] riscv: Kconfig: do not select PCI_MSI if CONIFG_PCI is enabled

[Jisheng Zhang]

On Tue, 20 Jul 2021 16:05:33 +0800
Vincent Chen <vincent.chen at sifive.com> wrote:

> Just a gentle ping.
> Does anyone have any comments on this RFC patch? If not, I will remove
> the "note" part from the git description and resend it as a formal
> patch. Thank you.
> 
> On Fri, Jul 9, 2021 at 11:12 AM Vincent Chen <vincent.chen at sifive.com> wrote:
> >
> > The CONFIG_PCI_MSI is used to allow device drivers to enable MSI. The MSI
> > enables a device to generate an interrupt using an inbound Memory Write
> > on its PCI bus instead of asserting a device IRQ pin. The whole mechanism
> > needs support from the PCI controller or generic interrupt controller and
> > the corresponding software driver.
> >
> > The RISC-V Kconfig file actively selects the PCI_MSI  if users enable
> > CONFIG_PCI. However, the RISC-V specification does not require every RISC-V
> > platform shall have MSI support. In other words, Kconfig enables CONFIG_PCI
> > to allow PCI devices to use MSI, but due to lack of MSI support, the kernel
> > may not have any function to deal with the MSI from PCI devices. When this
> > case happens, it leads to the following warning message displayed in
> > booting a v5.10~v5.12 kernel.
> >
> > ------------[ cut here ]------------
> > WARNING: CPU: 0 PID: 1 at include/linux/msi.h:219 __pci_enable_msix_range+0x4b6/0x50e
> > Modules linked in:
> > CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.10.34 #76
> > epc: ffffffe0004fa1e0 ra : ffffffe0004fa02e sp : ffffffe07fea7920
> >  gp : ffffffe0010e25c0 tp : ffffffe07fea8000 t0 : ffffffe0808c3180
> >  t1 : ffffffd000239000 t2 : 0000000000000400 s0 : ffffffe07fea79e0
> >  s1 : ffffffe080803000 a0 : 0000000000000000 a1 : ffffffe080803210
> >  a2 : ffffffe0808c30c0 a3 : 0000000000000101 a4 : ffffffe0808c30c0
> >  a5 : 0000000000000010 a6 : ffffffe0004dcf98 a7 : ffffffe07fe0f010
> >  s2 : 0000000000000002 s3 : 0000000000000014 s4 : 0000000000000000
> >  s5 : ffffffe080803210 s6 : 0000000000000000 s7 : ffffffe0808c3120
> >  s8 : ffffffe0808c30c0 s9 : 0000000000000000 s10: 0000000000000002
> >  s11: ffffffe0808030b0 t3 : 0000000000001000 t4 : 0000000000001000
> >  t5 : ffffffd040239000 t6 : ffffffe07ff8b2c0
> > status: 0000000000000120 badaddr: 0000000000000000 cause: 0000000000000003
> >
> > A simple reproduce way is to use defconfig to configure the 64-bit riscv
> > v5.10 ~ v5.12 kernel but disable CONFIG_PCIE_XILINX, and then booting this
> > kernel on the QEMU virt platform with virtio-mouse-pci.
> >
> > ARM's implementation may be a good sample for this case. Its Kconfig file
> > does not select CONFIG_PCI_MSI. Instead, It makes the selection of
> > CONFIG_PCI_MSI depends on the capability of the interrupt controller. This
> > way seems to be more straightforward. Therefore, this patch follows the
> > same way to remove the dependency between CONFIG_PCI and CONFIG_PCI_MSI
> > from Kconfig, which allows users to enable or disable CONFIG_PCI_MSI
> > according to the capabilities of the platform.
> >
> > (note) Why does this warning message only happen in the v5.10 to v5.12 kernel?
> > When the kernel wants to register the MSI ID of the first PCI device, if
> > the PCI device does not belong to any MSI domain and there is no valid
> > arch_setup_msi_irqs() defined by architecture or PCI controller, the
> > generic arch_setup_msi_irqs() will be called to display this warning
> > message. However, before v5.10, this flow is different. A valid and weak
> > arch_setup_msi_irqs() is defined in driver/pci/msi.c. The kernel can use
> > this arch_setup_msi_irqs() to set up the MSI. Therefore, it will not
> > encounter any problems even if the architecture and PCI controller does
> > not define it. After v5.13, the kernel uses the MSI domain to deal with
> > all the setup of MSI issues. Only some old PCI controller still uses
> > arch_setup_msi_irqs(). Through the hierarchy of the MSI domain, the kernel
> > can know the bus used by the PCI device has MSI support or not before
> > doing the MSI setup. In this condition, the kernel will return early before
> > calling the generic arch_setup_msi_irqs(), which will trigger the warning
> > message.
> >
> > Signed-off-by: Vincent Chen <vincent.chen at sifive.com>

Reviewed-by: Jisheng Zhang <jszhang at kernel.org>

> > ---
> >  arch/riscv/Kconfig | 1 -
> >  1 file changed, 1 deletion(-)
> >
> > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> > index 469a70bd8da6..563d550cb682 100644
> > --- a/arch/riscv/Kconfig
> > +++ b/arch/riscv/Kconfig
> > @@ -97,7 +97,6 @@ config RISCV
> >         select OF_EARLY_FLATTREE
> >         select OF_IRQ
> >         select PCI_DOMAINS_GENERIC if PCI
> > -       select PCI_MSI if PCI
> >         select RISCV_INTC
> >         select RISCV_TIMER if RISCV_SBI
> >         select SPARSE_IRQ
> > --
> > 2.7.4
> >  
> 
> _______________________________________________
> linux-riscv mailing list
> linux-riscv at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv