Re: [PATCH v3 4/4] riscv: Improve flush_tlb_kernel_range()
From: Alexandre Ghiti
Date: Fri Sep 08 2023 - 08:34:36 EST
Hi Prabhakar,
On Thu, Sep 7, 2023 at 12:50 PM Lad, Prabhakar
<prabhakar.csengg@xxxxxxxxx> wrote:
>
> Hi Alexandre,
>
> On Thu, Sep 7, 2023 at 10:06 AM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> >
> > Hi Prabhakar,
> >
> > On Wed, Sep 6, 2023 at 3:55 PM Lad, Prabhakar
> > <prabhakar.csengg@xxxxxxxxx> wrote:
> > >
> > > Hi Alexandre,
> > >
> > > On Wed, Sep 6, 2023 at 1:43 PM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> > > >
> > > > On Wed, Sep 6, 2023 at 2:24 PM Lad, Prabhakar
> > > > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > >
> > > > > Hi Alexandre,
> > > > >
> > > > > On Wed, Sep 6, 2023 at 1:18 PM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Wed, Sep 6, 2023 at 2:09 PM Lad, Prabhakar
> > > > > > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > > > >
> > > > > > > Hi Alexandre,
> > > > > > >
> > > > > > > On Wed, Sep 6, 2023 at 1:01 PM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> > > > > > > >
> > > > > > > > Hi Prabhakar,
> > > > > > > >
> > > > > > > > On Wed, Sep 6, 2023 at 1:49 PM Lad, Prabhakar
> > > > > > > > <prabhakar.csengg@xxxxxxxxx> wrote:
> > > > > > > > >
> > > > > > > > > Hi Alexandre,
> > > > > > > > >
> > > > > > > > > On Tue, Aug 1, 2023 at 9:58 AM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> > > > > > > > > >
> > > > > > > > > > This function used to simply flush the whole tlb of all harts, be more
> > > > > > > > > > subtile and try to only flush the range.
> > > > > > > > > >
> > > > > > > > > > The problem is that we can only use PAGE_SIZE as stride since we don't know
> > > > > > > > > > the size of the underlying mapping and then this function will be improved
> > > > > > > > > > only if the size of the region to flush is < threshold * PAGE_SIZE.
> > > > > > > > > >
> > > > > > > > > > Signed-off-by: Alexandre Ghiti <alexghiti@xxxxxxxxxxxx>
> > > > > > > > > > Reviewed-by: Andrew Jones <ajones@xxxxxxxxxxxxxxxx>
> > > > > > > > > > ---
> > > > > > > > > > arch/riscv/include/asm/tlbflush.h | 11 +++++-----
> > > > > > > > > > arch/riscv/mm/tlbflush.c | 34 +++++++++++++++++++++++--------
> > > > > > > > > > 2 files changed, 31 insertions(+), 14 deletions(-)
> > > > > > > > > >
> > > > > > > > > After applying this patch, I am seeing module load issues on RZ/Five
> > > > > > > > > (complete log [0]). I am testing defconfig + [1] (rz/five related
> > > > > > > > > configs).
> > > > > > > > >
> > > > > > > > > Any pointers on what could be an issue here?
> > > > > > > >
> > > > > > > > Can you give me the exact version of the kernel you use? The trap
> > > > > > > > addresses are vmalloc addresses, and a fix for those landed very late
> > > > > > > > in the release cycle.
> > > > > > > >
> > > > > > > I am using next-20230906, Ive pushed a branch [1] for you to have a look.
> > > > > > >
> > > > > > > [0] https://github.com/prabhakarlad/linux/tree/rzfive-debug
> > > > > >
> > > > > > Great, thanks, I had to get rid of this possibility :)
> > > > > >
> > > > > > As-is, I have no idea, can you try to "bisect" the problem? I mean
> > > > > > which patch in the series leads to those traps?
> > > > > >
> > > > > Oops sorry for not mentioning earlier, this is the offending patch
> > > > > which leads to the issues seen on rz/five.
> > > >
> > > > Ok, so at least I found the following problem, but I don't see how
> > > > that could fix your issue: can you give a try anyway? I keep looking
> > > > into this, thanks
> > > >
> > > > diff --git a/arch/riscv/mm/tlbflush.c b/arch/riscv/mm/tlbflush.c
> > > > index df2a0838c3a1..b5692bc6c76a 100644
> > > > --- a/arch/riscv/mm/tlbflush.c
> > > > +++ b/arch/riscv/mm/tlbflush.c
> > > > @@ -239,7 +239,7 @@ void flush_tlb_range(struct vm_area_struct *vma,
> > > > unsigned long start,
> > > >
> > > > void flush_tlb_kernel_range(unsigned long start, unsigned long end)
> > > > {
> > > > - __flush_tlb_range(NULL, start, end, PAGE_SIZE);
> > > > + __flush_tlb_range(NULL, start, end - start, PAGE_SIZE);
> > > > }
> > > >
> > > I am able to reproduce the issue with the above change too.
> >
> > I can't reproduce the problem on my Unmatched or Qemu, so it is not
> > easy to debug. But I took another look at your traces and something is
> > weird to me. In the following trace (and there is another one), the
> > trap is taken at 0xffffffff015ca034, which is the beginning of
> > rz_ssi_probe(): that's a page fault, so no translation was found (or
> > an invalid one is cached).
> >
> > [ 16.586527] Unable to handle kernel paging request at virtual
> > address ffffffff015ca034
> > [ 16.594750] Oops [#3]
> > ...
> > [ 16.622000] epc : rz_ssi_probe+0x0/0x52a [snd_soc_rz_ssi]
> > ...
> > [ 16.708697] status: 0000000200000120 badaddr: ffffffff015ca034
> > cause: 000000000000000c
> > [ 16.716580] [<ffffffff015ca034>] rz_ssi_probe+0x0/0x52a
> > [snd_soc_rz_ssi]
> > ...
> >
> > But then here we are able to read the code at this same address:
> > [ 16.821620] Code: 0109 6597 0000 8593 5f65 7097 7f34 80e7 7aa0 b7a9
> > (7139) f822
> >
> > So that looks like a "transient" error. Do you know if you uarch
> > caches invalid TLB entries? If you don't know, I have just written
> > some piece of code to determine if it does, let me know.
> >
> No I dont, can you please share the details so that I can pass on the
> information to you.
>
> > Do those errors always happen?
> >
> Yes they do.
>
I still can't reproduce those errors, I built different configs
including yours, insmod/rmmod a few modules but still can't reproduce
that. I'm having a hard time understanding how the correct mapping
magically appears in the trap handler. We finally removed this
patchset from 6.6...
You can give the following patch a try to determine if your uarch
caches invalid TLB entries, but honestly, I'm not sure if that would
help (but it will test my patch :)). The output can be seen in dmesg
"uarch caches invalid entries:".
If the trap addresses are constant, I would try to breakpoint on
flush_tlb_kernel_range() on those addresses and see what happens:
maybe that's an alignment issue or something else, maybe that's not
even called before the trap...etc. More info are welcome :)
Thanks!
diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index 80af436c04ac..8f863b251898 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -58,6 +58,8 @@ bool pgtable_l5_enabled = IS_ENABLED(CONFIG_64BIT)
&& !IS_ENABLED(CONFIG_XIP_KER
EXPORT_SYMBOL(pgtable_l4_enabled);
EXPORT_SYMBOL(pgtable_l5_enabled);
+bool tlb_caching_invalid_entries;
+
phys_addr_t phys_ram_base __ro_after_init;
EXPORT_SYMBOL(phys_ram_base);
@@ -752,6 +754,18 @@ static void __init disable_pgtable_l4(void)
satp_mode = SATP_MODE_39;
}
+static void __init enable_pgtable_l5(void)
+{
+ pgtable_l5_enabled = true;
+ satp_mode = SATP_MODE_57;
+}
+
+static void __init enable_pgtable_l4(void)
+{
+ pgtable_l4_enabled = true;
+ satp_mode = SATP_MODE_48;
+}
+
static int __init print_no4lvl(char *p)
{
pr_info("Disabled 4-level and 5-level paging");
@@ -828,6 +842,113 @@ static __init void set_satp_mode(uintptr_t dtb_pa)
memset(early_pud, 0, PAGE_SIZE);
memset(early_pmd, 0, PAGE_SIZE);
}
+
+/* Determine at runtime if the uarch caches invalid TLB entries */
+static __init void set_tlb_caching_invalid_entries(void)
+{
+#define NR_RETRIES_CACHING_INVALID_ENTRIES 50
+ uintptr_t set_tlb_caching_invalid_entries_pmd = ((unsigned
long)set_tlb_caching_invalid_entries) & PMD_MASK;
+ // TODO the test_addr as defined below could go into another pud...
+ uintptr_t test_addr = set_tlb_caching_invalid_entries_pmd + 2
* PMD_SIZE;
+ pmd_t valid_pmd;
+ u64 satp;
+ int i = 0;
+
+ /* To ease the page table creation */
+ // TODO use variable instead, like in the clean, nop stap_mode too
+ disable_pgtable_l5();
+ disable_pgtable_l4();
+
+ /* Establish a mapping for set_tlb_caching_invalid_entries() in sv39 */
+ create_pgd_mapping(early_pg_dir,
+ set_tlb_caching_invalid_entries_pmd,
+ (uintptr_t)early_pmd,
+ PGDIR_SIZE, PAGE_TABLE);
+
+ /* Handle the case where set_tlb_caching_invalid_entries
straddles 2 PMDs */
+ create_pmd_mapping(early_pmd,
+ set_tlb_caching_invalid_entries_pmd,
+ set_tlb_caching_invalid_entries_pmd,
+ PMD_SIZE, PAGE_KERNEL_EXEC);
+ create_pmd_mapping(early_pmd,
+ set_tlb_caching_invalid_entries_pmd + PMD_SIZE,
+ set_tlb_caching_invalid_entries_pmd + PMD_SIZE,
+ PMD_SIZE, PAGE_KERNEL_EXEC);
+
+ /* Establish an invalid mapping */
+ create_pmd_mapping(early_pmd, test_addr, 0, PMD_SIZE, __pgprot(0));
+
+ /* Precompute the valid pmd here because the mapping for
pfn_pmd() won't exist */
+ valid_pmd =
pfn_pmd(PFN_DOWN(set_tlb_caching_invalid_entries_pmd), PAGE_KERNEL);
+
+ local_flush_tlb_all();
+ satp = PFN_DOWN((uintptr_t)&early_pg_dir) | SATP_MODE_39;
+ csr_write(CSR_SATP, satp);
+
+ /*
+ * Set stvec to after the trapping access, access this invalid mapping
+ * and legitimately trap
+ */
+ // TODO: Should I save the previous stvec?
+#define ASM_STR(x) __ASM_STR(x)
+ asm volatile(
+ "la a0, 1f \n"
+ "csrw " ASM_STR(CSR_TVEC) ", a0 \n"
+ "ld a0, 0(%0) \n"
+ ".align 2 \n"
+ "1: \n"
+ :
+ : "r" (test_addr)
+ : "a0"
+ );
+
+ /* Now establish a valid mapping to check if the invalid one
is cached */
+ early_pmd[pmd_index(test_addr)] = valid_pmd;
+
+ /*
+ * Access the valid mapping multiple times: indeed, we can't use
+ * sfence.vma as a barrier to make sure the cpu did not reorder accesses
+ * so we may trap even if the uarch does not cache invalid entries. By
+ * trying a few times, we make sure that those uarchs will see the right
+ * mapping at some point.
+ */
+
+ i = NR_RETRIES_CACHING_INVALID_ENTRIES;
+
+#define ASM_STR(x) __ASM_STR(x)
+ asm_volatile_goto(
+ "la a0, 1f \n"
+ "csrw " ASM_STR(CSR_TVEC) ", a0 \n"
+ ".align 2 \n"
+ "1: \n"
+ "addi %0, %0, -1 \n"
+ "blt %0, zero, %l[caching_invalid_entries] \n"
+ "ld a0, 0(%1) \n"
+ :
+ : "r" (i), "r" (test_addr)
+ : "a0"
+ : caching_invalid_entries
+ );
+
+ csr_write(CSR_SATP, 0ULL);
+ local_flush_tlb_all();
+
+ /* If we don't trap, the uarch does not cache invalid entries! */
+ tlb_caching_invalid_entries = false;
+ goto clean;
+
+caching_invalid_entries:
+ csr_write(CSR_SATP, 0ULL);
+ local_flush_tlb_all();
+
+ tlb_caching_invalid_entries = true;
+clean:
+ memset(early_pg_dir, 0, PAGE_SIZE);
+ memset(early_pmd, 0, PAGE_SIZE);
+
+ enable_pgtable_l4();
+ enable_pgtable_l5();
+}
#endif
/*
@@ -1040,6 +1161,7 @@ asmlinkage void __init setup_vm(uintptr_t dtb_pa)
#endif
#if defined(CONFIG_64BIT) && !defined(CONFIG_XIP_KERNEL)
+ set_tlb_caching_invalid_entries();
set_satp_mode(dtb_pa);
#endif
@@ -1290,6 +1412,9 @@ static void __init setup_vm_final(void)
local_flush_tlb_all();
pt_ops_set_late();
+
+ pr_info("uarch caches invalid entries: %s",
+ tlb_caching_invalid_entries ? "yes": "no");
}
#else
asmlinkage void __init setup_vm(uintptr_t dtb_pa)
> Cheers,
> Prabhakar