From: David Hildenbrand <david@redhat.com>
To: Richard Henderson <richard.henderson@linaro.org>, qemu-devel@nongnu.org
Cc: Florian Weimer <fweimer@redhat.com>,
Thomas Huth <thuth@redhat.com>, Cornelia Huck <cohuck@redhat.com>,
Stefano Brivio <sbrivio@redhat.com>,
qemu-s390x@nongnu.org, Richard Henderson <rth@twiddle.net>
Subject: Re: [Qemu-devel] [PATCH v1 2/4] s390x/tcg: Introduce probe_read_access()
Date: Wed, 21 Aug 2019 20:48:39 +0200 [thread overview]
Message-ID: <2d73e310-a5d3-3cc1-0c84-c056ad942c00@redhat.com> (raw)
In-Reply-To: <dc42887a-f991-1fe0-36a7-a7804bc4a939@linaro.org>
On 21.08.19 19:26, Richard Henderson wrote:
> On 8/21/19 2:22 AM, David Hildenbrand wrote:
>> +/*
>> + * Make sure the read access is permitted and TLB entries are created. In
>> + * very rare cases it might happen that the actual accesses might need
>> + * new MMU translations. If the page tables were changed in between, we
>> + * might still trigger a fault. However, this seems to barely happen, so we
>> + * can ignore this for now.
>> + */
>> +void probe_read_access(CPUS390XState *env, uint64_t addr, uint64_t len,
>> + uintptr_t ra)
>> +{
>> +#ifdef CONFIG_USER_ONLY
>> + if (!guest_addr_valid(addr) || !guest_addr_valid(addr + len - 1) ||
>> + page_check_range(addr, len, PAGE_READ) < 0) {
>> + s390_program_interrupt(env, PGM_ADDRESSING, ILEN_AUTO, ra);
>> + }
>> +#else
>> + while (len) {
>> + const uint64_t pagelen = -(addr | -TARGET_PAGE_MASK);
>> + const uint64_t curlen = MIN(pagelen, len);
>> +
>> + cpu_ldub_data_ra(env, addr, ra);
>> + addr = wrap_address(env, addr + curlen);
>> + len -= curlen;
>> + }
>> +#endif
>> +}
>
> I don't think this is really the right approach, precisely because of the
> comment above.
>
> I think we should
>
> (1) Modify the generic probe_write to return the host address,
> akin to tlb_vaddr_to_host except it *will* fault.
>
> (2) Create a generic version of probe_write for CONFIG_USER_ONLY,
> much like the one you have done for target/s390x.
>
> (3) Create generic version of probe_read that does the same.
>
> (4) Rewrite fast_memset and fast_memmove to fetch all of the host
> addresses before doing any modifications. The functions are
> currently written as if len can be very large, handling any
> number of pages. Except that's not true. While there are
> several kinds of users apart from MVC, two pages are sufficient
> for all users.
>
> Well, should be. We would need to adjust do_mvcl to limit the
> operation to TARGET_PAGE_SIZE (CC=3, cpu-determined number of
> bytes moved without reaching end of first operand).
> Which is probably a good idea anyway. System mode should not
> spend forever executing one instruction, as it would if you
> pass in a 64-bit length from MVCLE.
>
Related to that: yes, that's what I mentioned in the cover letter, the
MOVE variants are full of issues. MVCLE should be limited to 4096 bytes,
and we should return cc=3. However, MVCL (which also uses do_mvcl) is
also semi-broken (it's an interruptible instruction - not cc=3, we have
to update the register step by step). MVCL should return cc=3 in case
it's an destructive copy - also not implemented properly. mem_helpers.c
is full of issues.
--
Thanks,
David / dhildenb
next prev parent reply other threads:[~2019-08-21 18:49 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-21 9:22 [Qemu-devel] [PATCH v1 0/4] s390x/tcg: MOVE (MVC): Fault-safe handling David Hildenbrand
2019-08-21 9:22 ` [Qemu-devel] [PATCH v1 1/4] s390x/tcg: Use guest_addr_valid() instead of h2g_valid() in probe_write_access() David Hildenbrand
2019-08-21 9:22 ` [Qemu-devel] [PATCH v1 2/4] s390x/tcg: Introduce probe_read_access() David Hildenbrand
2019-08-21 17:26 ` Richard Henderson
2019-08-21 17:37 ` David Hildenbrand
2019-08-21 19:19 ` Richard Henderson
2019-08-21 19:36 ` David Hildenbrand
2019-08-21 20:38 ` Richard Henderson
2019-08-21 21:33 ` David Hildenbrand
2019-08-21 22:31 ` Richard Henderson
2019-08-21 22:43 ` Richard Henderson
2019-08-22 6:42 ` David Hildenbrand
2019-08-22 7:01 ` David Hildenbrand
2019-08-26 9:31 ` David Hildenbrand
2019-08-21 18:48 ` David Hildenbrand [this message]
2019-08-21 9:22 ` [Qemu-devel] [PATCH v1 3/4] s390x/tcg: MOVE (MVC): Increment the length once David Hildenbrand
2019-08-21 15:47 ` Richard Henderson
2019-08-21 9:22 ` [Qemu-devel] [PATCH v1 4/4] s390x/tcg: MOVE (MVC): Fault-safe handling David Hildenbrand
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=2d73e310-a5d3-3cc1-0c84-c056ad942c00@redhat.com \
--to=david@redhat.com \
--cc=cohuck@redhat.com \
--cc=fweimer@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=qemu-s390x@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=rth@twiddle.net \
--cc=sbrivio@redhat.com \
--cc=thuth@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).