Staging/xlnx/ad9084 cfir sparse mode support

[mhennerich]

PR Description

Two commits touching ad9088_parse_cfilt() (the parser behind the cfir_config sysfs bin attribute):

1. Fix CFIR parser bugs and document blob format
   - enable: and coeff_transfer: both claimed BIT(6) in read_mask, silently dropping one when both were present and running the enable-path with uninitialized locals. Moved enable: to BIT(8) and widened read_mask to u16.
   - gain: was read into a u32 via %d, breaking negative-dB values. Use a dedicated s32.
   - Bound %s conversions in enable: / selection_mode: (cfir + pfilt) to %15s.
   - Rewrote the kernel-doc to enumerate every directive, its value set, the read_mask bit, the Apollo API it feeds, and the post-parse programming order.
2. Add CFIR sparse-mode programming
   - The blob accepted sparse_filt_en but never called adi_apollo_cfir_sparse_coeff_sel_pgm() / adi_apollo_cfir_sparse_mem_sel_pgm(), so sparse mode silently used stale HW state.
   - When sparse_filt_en: 1 precedes the coefficient block, each tap line is now (I/Q columns unchanged, hsel 0..63 appended); exactly 16 taps required.
   - New optional sparse_mem_sel: directive (each 0..3) for explicit Dstore0/2/5/7 spacing.
   - Programming sequence mirrors the Apollo fullchip_sparse_cfir() reference.
   - kdoc documents the Dstore/hsel model and the UG §6389 hsel=15/31/47/63 limitation.

PR Type

• Bug fix (a change that fixes an issue)
• New feature (a change that adds new functionality)

[github-actions]

Automated review run: https://github.com/analogdevicesinc/linux/actions/runs/24449775287

---

This series (6dacbb616..3bc9ff877) fixes parser bugs in the AD9088 CFIR/PFILT filter config blobs and adds CFIR sparse-mode programming. Two of the three commits are correct; one has an ordering bug.

✅ 6dacbb616 — Fix CFIR parser bugs

All three fixes look correct:

• BIT(6) collision between enable: and coeff_transfer: → enable: moved to BIT(8), read_mask widened to u16 ✅
• Negative dB gain (-18/-12/-6) silently lost through u32 → dedicated s32 gain_db ✅
• Unbounded %s in enable: / selection_mode: → %15s ✅

⚠️ 5e42d5eb5 — Add CFIR sparse-mode programming

One ordering bug: the sparse coefficient count is validated after adi_apollo_cfir_coeff_pgm() has already been called with uninitialized array tail entries.

adi_apollo_cfir_coeff_pgm() always programs exactly 16 taps regardless of the len argument (verified in public/src/adi_apollo_cfir.c:66). If fewer than 16 <i> <q> <hsel> lines are provided in sparse mode, uninitialized stack garbage is written to hardware before the count check fires:

// ad9088.c — coeff_pgm runs FIRST, with uninitialized cfir_coeff_i/q[i..15]
ret = adi_apollo_cfir_coeff_pgm(..., cfir_coeff_i, cfir_coeff_q, i);
if (ret < 0)
    goto out1;

if (cfir_pgm.cfir_sparse_filt_en) {
    if (i != ADI_APOLLO_CFIR_COEFF_NUM) {  // ← too late
        ...
        goto out1;  // ← still returns 'size' (success) to userspace

Additionally, sscanf(line, "sparse_mem_sel: %i %i %i", &m0, &m1, &m2) uses %i with u32 * (UB for negative input per C standard); same for hsel. Should be %u/%u respectively.

A fixup patch is suggested (attached to this run):

fixup! iio: trx-rf: ad9088: Add CFIR sparse-mode programming

- Move sparse coefficient count check BEFORE adi_apollo_cfir_coeff_pgm()
  and use goto out (returns -EINVAL) instead of goto out1 (returns size).
- Fix sparse_mem_sel: scanner from %i to %u for unsigned m0/m1/m2.
- Fix hsel scanner from %i %i %i to %i %i %u for unsigned hsel.

Patch stored at: /tmp/tmp.rhzMn4knoq/0001-fixup-iio-trx-rf-ad9088-Add-CFIR-sparse-mode-program.patch

✅ 3bc9ff877 — Fix PFILT parser bugs

Both fixes are correct:

• pfilt_mask%u / bank_mask%u now scanned from pre-extracted substrings p/b instead of the full line ✅
• Missing dest: now returns -EINVAL (matching CFIR parser) ✅

CI Warnings

All clang_analyzer warnings (uninitialized val at lines 1573/1599, dead stores at 1564/1569/1582/1593) and coccicheck unneeded-semicolon warnings at lines 64/65 are pre-existing and not introduced by this series. The checkpatch subject-line warning for commit 5e42d5eb5 is a false positive — the subject describes the feature, not a tool.

Build verified clean with gcc_aarch64 and llvm_x86 for ad9088.c.

[mhennerich]

Fixed:

• Move sparse coefficient count check BEFORE adi_apollo_cfir_coeff_pgm()
  and use goto out (returns -EINVAL) instead of goto out1 (returns size).

Did not fix:

• Fix sparse_mem_sel: scanner from %i to %u for unsigned m0/m1/m2.
• Fix hsel scanner from %i %i %i to %i %i %u for unsigned hsel.

%i parses decimals as well as hex values - that's what we need.

[nunojsa]

Just one remark from me

[nunojsa]

Should we ignore errors? Same for all the abov e

[mhennerich]

I'll take a look