Re: [RFC PATCH 05/16] RISC-V: KVM: Implement VCPU interrupts and requests handling

From: Anup Patel <anup@brainfault.org>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: Anup Patel <Anup.Patel@wdc.com>,
	Palmer Dabbelt <palmer@sifive.com>,
	Paul Walmsley <paul.walmsley@sifive.com>,
	Radim K <rkrcmar@redhat.com>,
	Daniel Lezcano <daniel.lezcano@linaro.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Atish Patra <Atish.Patra@wdc.com>,
	Alistair Francis <Alistair.Francis@wdc.com>,
	Damien Le Moal <Damien.LeMoal@wdc.com>,
	Christoph Hellwig <hch@infradead.org>,
	"kvm@vger.kernel.org" <kvm@vger.kernel.org>,
	"linux-riscv@lists.infradead.org"
	<linux-riscv@lists.infradead.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [RFC PATCH 05/16] RISC-V: KVM: Implement VCPU interrupts and requests handling
Date: Tue, 30 Jul 2019 18:15:39 +0530	[thread overview]
Message-ID: <CAAhSdy3b-o6y1fsYi1iQcCN=9ZuC98TLCqjHCYAzOCx+N+_89w@mail.gmail.com> (raw)
In-Reply-To: <66c4e468-7a69-31e7-778b-228908f0e737@redhat.com>

On Tue, Jul 30, 2019 at 5:42 PM Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> On 30/07/19 14:00, Anup Patel wrote:
> > On Tue, Jul 30, 2019 at 4:47 PM Paolo Bonzini <pbonzini@redhat.com> wrote:
> >>
> >> First, something that is not clear to me: how do you deal with a guest
> >> writing 1 to VSIP.SSIP?  I think that could lead to lost interrupts if
> >> you have the following sequence
> >>
> >> 1) guest writes 1 to VSIP.SSIP
> >>
> >> 2) guest leaves VS-mode
> >>
> >> 3) host syncs VSIP
> >>
> >> 4) user mode triggers interrupt
> >>
> >> 5) host reenters guest
> >>
> >> 6) host moves irqs_pending to VSIP and clears VSIP.SSIP in the process
> >
> > This reasoning also apply to M-mode firmware (OpenSBI) providing timer
> > and IPI services to HS-mode software. We had some discussion around
> > it in a different context.
> > (Refer, https://github.com/riscv/opensbi/issues/128)
> >
> > The thing is SIP CSR is supposed to be read-only for any S-mode SW. This
> > means HS-mode/VS-mode SW modifications to SIP CSR should have no
> > effect.
>
> Is it?  The privileged specification says
>
>   Interprocessor interrupts are sent to other harts by implementation-
>   specific means, which will ultimately cause the SSIP bit to be set in
>   the recipient hart’s sip register.

To further explain my rationale ...

Here's some text from RISC-V spec regarding MIP CSR:
"The mip register is an MXLEN-bit read/write register containing information
on pending interrupts, while mie is the corresponding MXLEN-bit read/write
register containing interrupt enable bits. Only the bits corresponding to
lower-privilege software interrupts (USIP, SSIP), timer interrupts (UTIP, STIP),
and external interrupts (UEIP, SEIP) in mip are writable through this CSR
address; the remaining bits are read-only."

Here's some text from RISC-V spec regarding SIP CSR:
"software interrupt-pending (SSIP) bit in the sip register. A pending
supervisor-level software interrupt can be cleared by writing 0 to the SSIP bit
in sip. Supervisor-level software interrupts are disabled when the SSIE bit in
the sie register is clear."

Without RISC-V hypervisor extension, the SIP is essentially a restricted
view of MIP CSR. Also as-per above, S-mode SW can only write 0 to SSIP
bit in SIP CSR whereas it can only be set by M-mode SW or some HW
mechanism (such as S-mode CLINT).

There was quite a bit of discussion in last RISC-V Zurich Workshop about
avoiding SBI calls for injecting IPIs. The best suggestion so far is to
eventually have RISC-V systems with separate CLINT HW for M-mode
and S-mode. The S-mode SW can use S-mode CLINT to trigger IPIs to
other CPUs and it will use SBI calls for IPIs only when S-mode CLINT is
not available.

>
>   All bits besides SSIP in the sip register are read-only.
>
> Meaning that sending an IPI to self by writing 1 to sip.SSIP is
> well-defined.  The same should be true of vsip.SSIP while in VS mode.
>
> > Do you still an issue here?
>
> Do you see any issues in the pseudocode I sent?  It gets away with the
> spinlock and request so it may be a good idea anyway. :)

Yes, I am evaluating your psedocode right now. I definitely want to
remove the irq_pending_lock if possible. I will try to in-corporate your
suggestion in v2 series.

Regards,
Anup