From: Pierrick Bouvier <pierrick.bouvier@linaro.org>
To: Richard Henderson <richard.henderson@linaro.org>, qemu-devel@nongnu.org
Cc: pbonzini@redhat.com, philmd@linaro.org
Subject: Re: [PATCH 09/37] accel/tcg: Perform aligned atomic reads in translator_ld
Date: Thu, 13 Mar 2025 10:14:31 -0700 [thread overview]
Message-ID: <51cfbbc5-c258-49af-bcab-ad1ce21ff711@linaro.org> (raw)
In-Reply-To: <20250313034524.3069690-10-richard.henderson@linaro.org>
On 3/12/25 20:44, Richard Henderson wrote:
> Perform aligned atomic reads in translator_ld, if possible.
> According to
>
> https://lore.kernel.org/qemu-devel/20240607101403.1109-1-jim.shu@sifive.com/
>
> this is required for RISC-V Ziccif.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
> accel/tcg/translator.c | 42 ++++++++++++++++++++++++++++++++++++++----
> 1 file changed, 38 insertions(+), 4 deletions(-)
>
> diff --git a/accel/tcg/translator.c b/accel/tcg/translator.c
> index ef1538b4fc..0260fb1915 100644
> --- a/accel/tcg/translator.c
> +++ b/accel/tcg/translator.c
> @@ -265,12 +265,14 @@ static bool translator_ld(CPUArchState *env, DisasContextBase *db,
>
> if (likely(((base ^ last) & TARGET_PAGE_MASK) == 0)) {
> /* Entire read is from the first page. */
> - memcpy(dest, host + (pc - base), len);
> - return true;
> + goto do_read;
> }
>
> if (unlikely(((base ^ pc) & TARGET_PAGE_MASK) == 0)) {
> - /* Read begins on the first page and extends to the second. */
> + /*
> + * Read begins on the first page and extends to the second.
> + * The unaligned read is never atomic.
> + */
> size_t len0 = -(pc | TARGET_PAGE_MASK);
> memcpy(dest, host + (pc - base), len0);
> pc += len0;
> @@ -329,7 +331,39 @@ static bool translator_ld(CPUArchState *env, DisasContextBase *db,
> host = db->host_addr[1];
> }
>
> - memcpy(dest, host + (pc - base), len);
> + do_read:
> + /*
> + * Assume aligned reads should be atomic, if possible.
> + * We're not in a position to jump out with EXCP_ATOMIC.
> + */
> + host += pc - base;
> + switch (len) {
> + case 2:
> + if (QEMU_IS_ALIGNED(pc, 2)) {
> + uint16_t t = qatomic_read((uint16_t *)host);
> + stw_he_p(dest, t);
> + return true;
> + }
> + break;
> + case 4:
> + if (QEMU_IS_ALIGNED(pc, 4)) {
> + uint32_t t = qatomic_read((uint32_t *)host);
> + stl_he_p(dest, t);
> + return true;
> + }
> + break;
> +#ifdef CONFIG_ATOMIC64
> + case 8:
> + if (QEMU_IS_ALIGNED(pc, 8)) {
> + uint64_t t = qatomic_read__nocheck((uint64_t *)host);
> + stl_he_p(dest, t);
Should it be stq_he_p?
> + return true;
> + }
> + break;
> +#endif
> + }
> + /* Unaligned or partial read from the second page is not atomic. */
> + memcpy(dest, host, len);
> return true;
> }
>
next prev parent reply other threads:[~2025-03-13 17:14 UTC|newest]
Thread overview: 115+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-03-13 3:44 [PATCH 00/37] accel/tcg, codebase: Build once patches Richard Henderson
2025-03-13 3:44 ` [PATCH 01/37] accel/tcg: Build user-exec-stub.c once Richard Henderson
2025-03-13 9:43 ` Philippe Mathieu-Daudé
2025-03-13 16:37 ` Pierrick Bouvier
2025-03-13 22:56 ` Alex Bennée
2025-03-13 3:44 ` [PATCH 02/37] accel/tcg: Build plugin-gen.c once Richard Henderson
2025-03-13 16:46 ` Pierrick Bouvier
2025-03-13 22:57 ` Alex Bennée
2025-03-13 3:44 ` [PATCH 03/37] include/exec: Use vaddr for *_mmu guest memory access routines Richard Henderson
2025-03-13 16:48 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 04/37] include/exec: Split out cpu-ldst-common.h Richard Henderson
2025-03-13 16:50 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 05/37] include/exec: Inline *_mmuidx_ra memory operations Richard Henderson
2025-03-13 16:59 ` Pierrick Bouvier
2025-03-13 18:05 ` Richard Henderson
2025-03-13 18:26 ` Philippe Mathieu-Daudé
2025-03-14 0:07 ` Richard Henderson
2025-03-13 20:14 ` Pierrick Bouvier
2025-03-13 17:11 ` Pierrick Bouvier
2025-03-13 17:55 ` Richard Henderson
2025-03-13 3:44 ` [PATCH 06/37] include/exec: Inline *_data_ra " Richard Henderson
2025-03-13 16:22 ` Alex Bennée
2025-03-14 0:20 ` Richard Henderson
2025-03-13 17:04 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 07/37] include/exec: Inline *_data " Richard Henderson
2025-03-13 17:05 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 08/37] include/exec: Inline *_code " Richard Henderson
2025-03-13 17:08 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 09/37] accel/tcg: Perform aligned atomic reads in translator_ld Richard Henderson
2025-03-13 17:14 ` Pierrick Bouvier [this message]
2025-03-14 0:47 ` Richard Henderson
2025-03-13 3:44 ` [PATCH 10/37] accel/tcg: Use cpu_ld*_code_mmu in translator.c Richard Henderson
2025-03-13 9:54 ` Philippe Mathieu-Daudé
2025-03-13 9:57 ` Philippe Mathieu-Daudé
2025-03-13 17:21 ` Pierrick Bouvier
2025-03-14 0:57 ` Richard Henderson
2025-03-13 3:44 ` [PATCH 11/37] accel/tcg: Implement translator_ld*_end Richard Henderson
2025-03-13 17:33 ` Pierrick Bouvier
2025-03-13 18:17 ` Richard Henderson
2025-03-13 20:49 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 12/37] accel/tcg: Remove mmap_lock/unlock from watchpoint.c Richard Henderson
2025-03-13 9:57 ` Philippe Mathieu-Daudé
2025-03-13 17:27 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 13/37] include/exec: Split out mmap-lock.h Richard Henderson
2025-03-13 9:58 ` Philippe Mathieu-Daudé
2025-03-14 1:40 ` Richard Henderson
2025-03-13 17:34 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 14/37] include/system: Move exec/memory.h to system/memory.h Richard Henderson
2025-03-13 10:00 ` Philippe Mathieu-Daudé
2025-03-13 17:34 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 15/37] include/system: Move exec/address-spaces.h to system/address-spaces.h Richard Henderson
2025-03-13 10:01 ` Philippe Mathieu-Daudé
2025-03-13 17:34 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 16/37] include/system: Move exec/ioport.h to system/ioport.h Richard Henderson
2025-03-13 17:35 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 17/37] meson: Introduce top-level libuser_ss and libsystem_ss Richard Henderson
2025-03-13 17:36 ` Pierrick Bouvier
2025-03-13 3:44 ` [PATCH 18/37] gdbstub: Move syscalls.c out of common_ss Richard Henderson
2025-03-13 17:46 ` Pierrick Bouvier
2025-03-13 18:31 ` Richard Henderson
2025-03-13 3:44 ` [PATCH 19/37] accel/tcg: Use libuser_ss and libsystem_ss Richard Henderson
2025-03-13 20:53 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 20/37] target/mips: Protect semihosting call with CONFIG_SEMIHOSTING Richard Henderson
2025-03-13 20:51 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 21/37] target/xtensa: " Richard Henderson
2025-03-13 20:54 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 22/37] semihosting: Move user-only implementation out-of-line Richard Henderson
2025-03-13 10:04 ` Philippe Mathieu-Daudé
2025-03-13 10:28 ` Philippe Mathieu-Daudé
2025-03-13 3:45 ` [PATCH 23/37] include/system: Move exec/ram_addr.h to system/ram_addr.h Richard Henderson
2025-03-13 17:22 ` Alex Bennée
2025-03-13 20:55 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 24/37] include/system: Move exec/ramblock.h to system/ramblock.h Richard Henderson
2025-03-13 20:56 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 25/37] common-user: Split out watchpoint-stub.c Richard Henderson
2025-03-13 10:07 ` Philippe Mathieu-Daudé
2025-03-13 10:39 ` Philippe Mathieu-Daudé
2025-03-14 16:37 ` Richard Henderson
2025-03-14 16:53 ` Pierrick Bouvier
2025-03-13 20:57 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 26/37] hw/core: Move unconditional files to libsystem_ss, libuser_ss Richard Henderson
2025-03-13 20:57 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 27/37] system: Move watchpoint.c to libsystem_ss Richard Henderson
2025-03-13 20:57 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 28/37] plugins: Move api.c, core.c to libuser_ss, libsystem_ss Richard Henderson
2025-03-13 20:57 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 29/37] include/exec: Split out cpu-mmu-index.h Richard Henderson
2025-03-13 20:58 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 30/37] include/exec: Drop ifndef CONFIG_USER_ONLY from cpu-common.h Richard Henderson
2025-03-13 20:58 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 31/37] include/hw/core: Drop ifndef CONFIG_USER_ONLY from cpu.h Richard Henderson
2025-03-13 20:58 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 32/37] include/hw/intc: Remove ifndef CONFIG_USER_ONLY from armv7m_nvic.h Richard Henderson
2025-03-13 21:00 ` Pierrick Bouvier
2025-03-14 18:13 ` Richard Henderson
2025-03-14 18:36 ` Pierrick Bouvier
2025-03-14 20:03 ` Richard Henderson
2025-03-14 20:34 ` Pierrick Bouvier
2025-03-14 20:59 ` Richard Henderson
2025-03-14 22:05 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 33/37] include/hw/s390x: Remove ifndef CONFIG_USER_ONLY in css.h Richard Henderson
2025-03-13 21:00 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 34/37] include/system: Remove ifndef CONFIG_USER_ONLY Richard Henderson
2025-03-13 10:18 ` Philippe Mathieu-Daudé
2025-03-13 21:02 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 35/37] include/qemu: Remove ifndef CONFIG_USER_ONLY from accel.h Richard Henderson
2025-03-13 21:03 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 36/37] target/riscv: Remove ifndef CONFIG_USER_ONLY from cpu_cfg.h Richard Henderson
2025-03-13 21:03 ` Pierrick Bouvier
2025-03-13 3:45 ` [PATCH 37/37] meson: Only allow CONFIG_USER_ONLY from certain source sets Richard Henderson
2025-03-13 20:50 ` Pierrick Bouvier
2025-03-13 13:56 ` [PATCH 00/37] accel/tcg, codebase: Build once patches Alex Bennée
2025-03-13 16:36 ` Pierrick Bouvier
2025-03-13 17:48 ` Philippe Mathieu-Daudé
2025-03-13 20:13 ` Pierrick Bouvier
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=51cfbbc5-c258-49af-bcab-ad1ce21ff711@linaro.org \
--to=pierrick.bouvier@linaro.org \
--cc=pbonzini@redhat.com \
--cc=philmd@linaro.org \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
/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).