From: Tony Nguyen <tony.nguyen@bt.com>
To: Richard Henderson <richard.henderson@linaro.org>
Cc: peter.maydell@linaro.org, mark.cave-ayland@ilande.co.uk,
qemu-devel@nongnu.org
Subject: Re: [Qemu-devel] [PATCH 3/3] cputlb: Introduce TLB_BSWAP
Date: Thu, 12 Sep 2019 00:56:29 +1000 [thread overview]
Message-ID: <20190911145629.GB42643@imac.local> (raw)
In-Reply-To: <20190911014353.5926-4-richard.henderson@linaro.org>
On Tue, Sep 10, 2019 at 09:43:53PM -0400, Richard Henderson wrote:
> Handle bswap on ram directly in load/store_helper. This fixes a
> bug with the previous implementation in that one cannot use the
> I/O path for RAM.
>
> Fixes: a26fc6f5152b47f1
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
> include/exec/cpu-all.h | 2 +
> accel/tcg/cputlb.c | 118 ++++++++++++++++++++---------------------
> 2 files changed, 59 insertions(+), 61 deletions(-)
>
> diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h
> index d2d443c4f9..3928edab9a 100644
> --- a/include/exec/cpu-all.h
> +++ b/include/exec/cpu-all.h
> @@ -331,6 +331,8 @@ CPUArchState *cpu_copy(CPUArchState *env);
> #define TLB_MMIO (1 << (TARGET_PAGE_BITS - 3))
> /* Set if TLB entry contains a watchpoint. */
> #define TLB_WATCHPOINT (1 << (TARGET_PAGE_BITS - 4))
> +/* Set if TLB entry requires byte swap. */
> +#define TLB_BSWAP (1 << (TARGET_PAGE_BITS - 5))
>
> /* Use this mask to check interception with an alignment mask
> * in a TCG backend.
> diff --git a/accel/tcg/cputlb.c b/accel/tcg/cputlb.c
> index e6229d100a..eace6c82e3 100644
> --- a/accel/tcg/cputlb.c
> +++ b/accel/tcg/cputlb.c
> @@ -737,8 +737,7 @@ void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
> address |= TLB_INVALID_MASK;
> }
> if (attrs.byte_swap) {
> - /* Force the access through the I/O slow path. */
> - address |= TLB_MMIO;
> + address |= TLB_BSWAP;
> }
> if (!memory_region_is_ram(section->mr) &&
> !memory_region_is_romd(section->mr)) {
> @@ -901,10 +900,6 @@ static uint64_t io_readx(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
> bool locked = false;
> MemTxResult r;
>
> - if (iotlbentry->attrs.byte_swap) {
> - op ^= MO_BSWAP;
> - }
> -
> section = iotlb_to_section(cpu, iotlbentry->addr, iotlbentry->attrs);
> mr = section->mr;
> mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
> @@ -947,10 +942,6 @@ static void io_writex(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
> bool locked = false;
> MemTxResult r;
>
> - if (iotlbentry->attrs.byte_swap) {
> - op ^= MO_BSWAP;
> - }
> -
> section = iotlb_to_section(cpu, iotlbentry->addr, iotlbentry->attrs);
> mr = section->mr;
> mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
> @@ -1322,7 +1313,8 @@ static inline uint64_t direct_ldul_le(const void *haddr)
> static inline uint64_t ALWAYS_INLINE
> load_helper(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
> uintptr_t retaddr, MemOp op, bool code_read,
> - FullLoadHelper *full_load, DirectLoadHelper *direct)
> + FullLoadHelper *full_load, DirectLoadHelper *direct,
> + DirectLoadHelper *direct_swap)
> {
> uintptr_t mmu_idx = get_mmuidx(oi);
> uintptr_t index = tlb_index(env, mmu_idx, addr);
> @@ -1372,26 +1364,27 @@ load_helper(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
> /* On watchpoint hit, this will longjmp out. */
> cpu_check_watchpoint(env_cpu(env), addr, size,
> iotlbentry->attrs, BP_MEM_READ, retaddr);
> -
> - /* The backing page may or may not require I/O. */
> - tlb_addr &= ~TLB_WATCHPOINT;
> - if ((tlb_addr & ~TARGET_PAGE_MASK) == 0) {
> - goto do_aligned_access;
> - }
> }
>
> /* Handle I/O access. */
> - return io_readx(env, iotlbentry, mmu_idx, addr,
> - retaddr, access_type, op);
> - }
> + if (likely(tlb_addr & TLB_MMIO)) {
> + return io_readx(env, iotlbentry, mmu_idx, addr,
> + retaddr, access_type,
> + op ^ (tlb_addr & TLB_BSWAP ? MO_BSWAP : 0));
> + }
Previously, the end of if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) branch
called and returned the result of io_readx.
io_readx calls cpu_transaction_failed if memory_region_dispatch_read does not
return MEMTX_OK.
Is the end of the branch missing a cpu_transaction_failed call?
Similar question for store_helper.
>
> - /* Handle slow unaligned access (it spans two pages or IO). */
> - if (size > 1
> - && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
> - >= TARGET_PAGE_SIZE)) {
> + if (unlikely(tlb_addr & TLB_BSWAP)) {
> + haddr = (void *)((uintptr_t)addr + entry->addend);
> + return direct_swap(haddr);
> + }
> + } else if (size > 1
> + && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
> + >= TARGET_PAGE_SIZE)) {
> + /* Handle slow unaligned access (it spans two pages or IO). */
> target_ulong addr1, addr2;
> uint64_t r1, r2;
> unsigned shift;
> +
> do_unaligned_access:
> addr1 = addr & ~((target_ulong)size - 1);
> addr2 = addr1 + size;
> @@ -1409,7 +1402,6 @@ load_helper(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
> return res & MAKE_64BIT_MASK(0, size * 8);
> }
>
> - do_aligned_access:
> haddr = (void *)((uintptr_t)addr + entry->addend);
> return direct(haddr);
> }
> @@ -1428,7 +1420,7 @@ static uint64_t full_ldub_mmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_UB, false,
> - full_ldub_mmu, direct_ldub);
> + full_ldub_mmu, direct_ldub, direct_ldub);
> }
>
> tcg_target_ulong helper_ret_ldub_mmu(CPUArchState *env, target_ulong addr,
> @@ -1441,7 +1433,7 @@ static uint64_t full_le_lduw_mmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_LEUW, false,
> - full_le_lduw_mmu, direct_lduw_le);
> + full_le_lduw_mmu, direct_lduw_le, direct_lduw_be);
> }
>
> tcg_target_ulong helper_le_lduw_mmu(CPUArchState *env, target_ulong addr,
> @@ -1454,7 +1446,7 @@ static uint64_t full_be_lduw_mmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_BEUW, false,
> - full_be_lduw_mmu, direct_lduw_be);
> + full_be_lduw_mmu, direct_lduw_be, direct_lduw_le);
> }
>
> tcg_target_ulong helper_be_lduw_mmu(CPUArchState *env, target_ulong addr,
> @@ -1467,7 +1459,7 @@ static uint64_t full_le_ldul_mmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_LEUL, false,
> - full_le_ldul_mmu, direct_ldul_le);
> + full_le_ldul_mmu, direct_ldul_le, direct_ldul_be);
> }
>
> tcg_target_ulong helper_le_ldul_mmu(CPUArchState *env, target_ulong addr,
> @@ -1480,7 +1472,7 @@ static uint64_t full_be_ldul_mmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_BEUL, false,
> - full_be_ldul_mmu, direct_ldul_be);
> + full_be_ldul_mmu, direct_ldul_be, direct_ldul_le);
> }
>
> tcg_target_ulong helper_be_ldul_mmu(CPUArchState *env, target_ulong addr,
> @@ -1493,14 +1485,14 @@ uint64_t helper_le_ldq_mmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_LEQ, false,
> - helper_le_ldq_mmu, ldq_le_p);
> + helper_le_ldq_mmu, ldq_le_p, ldq_be_p);
> }
>
> uint64_t helper_be_ldq_mmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_BEQ, false,
> - helper_be_ldq_mmu, ldq_be_p);
> + helper_be_ldq_mmu, ldq_be_p, ldq_le_p);
> }
>
> /*
> @@ -1573,7 +1565,7 @@ static inline void direct_stl_le(void *haddr, uint64_t val)
> static inline void ALWAYS_INLINE
> store_helper(CPUArchState *env, target_ulong addr, uint64_t val,
> TCGMemOpIdx oi, uintptr_t retaddr, MemOp op,
> - DirectStoreHelper *direct)
> + DirectStoreHelper *direct, DirectStoreHelper *direct_swap)
> {
> uintptr_t mmu_idx = get_mmuidx(oi);
> uintptr_t index = tlb_index(env, mmu_idx, addr);
> @@ -1618,23 +1610,24 @@ store_helper(CPUArchState *env, target_ulong addr, uint64_t val,
> /* On watchpoint hit, this will longjmp out. */
> cpu_check_watchpoint(env_cpu(env), addr, size,
> iotlbentry->attrs, BP_MEM_WRITE, retaddr);
> -
> - /* The backing page may or may not require I/O. */
> - tlb_addr &= ~TLB_WATCHPOINT;
> - if ((tlb_addr & ~TARGET_PAGE_MASK) == 0) {
> - goto do_aligned_access;
> - }
> }
>
> /* Handle I/O access. */
> - io_writex(env, iotlbentry, mmu_idx, val, addr, retaddr, op);
> - return;
> - }
> + if (likely(tlb_addr & TLB_MMIO)) {
> + io_writex(env, iotlbentry, mmu_idx, val, addr, retaddr,
> + op ^ (tlb_addr & TLB_BSWAP ? MO_BSWAP : 0));
> + return;
> + }
>
> - /* Handle slow unaligned access (it spans two pages or IO). */
> - if (size > 1
> - && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
> - >= TARGET_PAGE_SIZE)) {
> + if (unlikely(tlb_addr & TLB_BSWAP)) {
> + haddr = (void *)((uintptr_t)addr + entry->addend);
> + direct_swap(haddr, val);
> + return;
> + }
> + } else if (size > 1
> + && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
> + >= TARGET_PAGE_SIZE)) {
> + /* Handle slow unaligned access (it spans two pages or IO). */
> int i;
> uintptr_t index2;
> CPUTLBEntry *entry2;
> @@ -1696,7 +1689,6 @@ store_helper(CPUArchState *env, target_ulong addr, uint64_t val,
> return;
> }
>
> - do_aligned_access:
> haddr = (void *)((uintptr_t)addr + entry->addend);
> direct(haddr, val);
> }
> @@ -1704,43 +1696,47 @@ store_helper(CPUArchState *env, target_ulong addr, uint64_t val,
> void helper_ret_stb_mmu(CPUArchState *env, target_ulong addr, uint8_t val,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> - store_helper(env, addr, val, oi, retaddr, MO_UB, direct_stb);
> + store_helper(env, addr, val, oi, retaddr, MO_UB, direct_stb, direct_stb);
> }
>
> void helper_le_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> - store_helper(env, addr, val, oi, retaddr, MO_LEUW, direct_stw_le);
> + store_helper(env, addr, val, oi, retaddr, MO_LEUW,
> + direct_stw_le, direct_stw_be);
> }
>
> void helper_be_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> - store_helper(env, addr, val, oi, retaddr, MO_BEUW, direct_stw_be);
> + store_helper(env, addr, val, oi, retaddr, MO_BEUW,
> + direct_stw_be, direct_stw_le);
> }
>
> void helper_le_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> - store_helper(env, addr, val, oi, retaddr, MO_LEUL, direct_stl_le);
> + store_helper(env, addr, val, oi, retaddr, MO_LEUL,
> + direct_stl_le, direct_stl_be);
> }
>
> void helper_be_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> - store_helper(env, addr, val, oi, retaddr, MO_BEUL, direct_stl_be);
> + store_helper(env, addr, val, oi, retaddr, MO_BEUL,
> + direct_stl_be, direct_stl_le);
> }
>
> void helper_le_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> - store_helper(env, addr, val, oi, retaddr, MO_LEQ, stq_le_p);
> + store_helper(env, addr, val, oi, retaddr, MO_LEQ, stq_le_p, stq_be_p);
> }
>
> void helper_be_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> - store_helper(env, addr, val, oi, retaddr, MO_BEQ, stq_be_p);
> + store_helper(env, addr, val, oi, retaddr, MO_BEQ, stq_be_p, stq_le_p);
> }
>
> /* First set of helpers allows passing in of OI and RETADDR. This makes
> @@ -1806,7 +1802,7 @@ static uint64_t full_ldub_cmmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_8, true,
> - full_ldub_cmmu, direct_ldub);
> + full_ldub_cmmu, direct_ldub, direct_ldub);
> }
>
> uint8_t helper_ret_ldb_cmmu(CPUArchState *env, target_ulong addr,
> @@ -1819,7 +1815,7 @@ static uint64_t full_le_lduw_cmmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_LEUW, true,
> - full_le_lduw_cmmu, direct_lduw_le);
> + full_le_lduw_cmmu, direct_lduw_le, direct_lduw_be);
> }
>
> uint16_t helper_le_ldw_cmmu(CPUArchState *env, target_ulong addr,
> @@ -1832,7 +1828,7 @@ static uint64_t full_be_lduw_cmmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_BEUW, true,
> - full_be_lduw_cmmu, direct_lduw_be);
> + full_be_lduw_cmmu, direct_lduw_be, direct_lduw_le);
> }
>
> uint16_t helper_be_ldw_cmmu(CPUArchState *env, target_ulong addr,
> @@ -1845,7 +1841,7 @@ static uint64_t full_le_ldul_cmmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_LEUL, true,
> - full_le_ldul_cmmu, direct_ldul_le);
> + full_le_ldul_cmmu, direct_ldul_le, direct_ldul_be);
> }
>
> uint32_t helper_le_ldl_cmmu(CPUArchState *env, target_ulong addr,
> @@ -1858,7 +1854,7 @@ static uint64_t full_be_ldul_cmmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_BEUL, true,
> - full_be_ldul_cmmu, direct_ldul_be);
> + full_be_ldul_cmmu, direct_ldul_be, direct_ldul_le);
> }
>
> uint32_t helper_be_ldl_cmmu(CPUArchState *env, target_ulong addr,
> @@ -1871,12 +1867,12 @@ uint64_t helper_le_ldq_cmmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_LEQ, true,
> - helper_le_ldq_cmmu, ldq_le_p);
> + helper_le_ldq_cmmu, ldq_le_p, ldq_be_p);
> }
>
> uint64_t helper_be_ldq_cmmu(CPUArchState *env, target_ulong addr,
> TCGMemOpIdx oi, uintptr_t retaddr)
> {
> return load_helper(env, addr, oi, retaddr, MO_BEQ, true,
> - helper_be_ldq_cmmu, ldq_be_p);
> + helper_be_ldq_cmmu, ldq_be_p, ldq_le_p);
> }
> --
> 2.17.1
>
>
next prev parent reply other threads:[~2019-09-11 15:17 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-11 1:43 [Qemu-devel] [PATCH 0/3] cputlb: Adjust tlb bswap implementation Richard Henderson
2019-09-11 1:43 ` [Qemu-devel] [PATCH 1/3] cputlb: Disable __always_inline__ without optimization Richard Henderson
2019-09-11 8:34 ` Peter Maydell
2019-09-11 1:43 ` [Qemu-devel] [PATCH 2/3] cputlb: Replace switches in load/store_helper with callback Richard Henderson
2019-09-11 10:55 ` Tony Nguyen
2019-09-11 13:22 ` Richard Henderson
2019-09-11 15:17 ` Tony Nguyen
2019-09-11 13:07 ` Philippe Mathieu-Daudé
2019-09-11 13:20 ` Richard Henderson
2019-09-11 1:43 ` [Qemu-devel] [PATCH 3/3] cputlb: Introduce TLB_BSWAP Richard Henderson
2019-09-11 14:56 ` Tony Nguyen [this message]
2019-09-12 19:07 ` Richard Henderson
2019-09-11 3:46 ` [Qemu-devel] [PATCH 0/3] cputlb: Adjust tlb bswap implementation no-reply
2019-09-11 9:55 ` no-reply
2019-09-11 13:07 ` no-reply
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=20190911145629.GB42643@imac.local \
--to=tony.nguyen@bt.com \
--cc=mark.cave-ayland@ilande.co.uk \
--cc=peter.maydell@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).